Tech News

Countries around the world are seeking to exert more control over content on the internet – and, by extension, their citizens. With more acts of terrorism taking place everywhere, they are now attaining a kind of online history too. With material like those from the recent Christchurch shooting proliferate as supporters upload them to any media platform they can reach. And lawmakers around the world have had enough, so this year, they hope to enact new legislation that will hold big tech companies like Facebook and Google more accountable for any terrorist-related content they host. The Australian parliament passed legislation to crack down on violent videos on social media. Recently Sen. Elizabeth Warren, US 2020 presidential hopeful proposed to build strong anti-trust laws and break big tech companies like Amazon, Google, Facebook and Apple. And on 3rd April, Elizabeth introduced Corporate Executive Accountability Act, a new piece of legislation that would make it easier to criminally charge company executives when Americans’ personal data is breached. Another news from Washington post states that UK has drafted an aggressive new plan to penalise Facebook, Google and other tech giants that don't stop the spread of harmful content online. Last year, the German parliament enacted the NetzDG law, requiring large social media sites to remove posts that violate certain provisions of the German code, including broad prohibitions on “defamation of religion,” “hate speech,” and “insult.” The removal obligation is triggered not by a court order, but by complaints from users. Companies must remove the posts within 24 hours or seven days facing steep fines if they fail to do so. Joining the bandwagon, Europe has also drafted EU Regulation on preventing the dissemination of terrorist content online. The legislation was first proposed by the EU last September as a response to the spread of ISIS propaganda online which encouraged further attacks. It covers recruiting materials such as displays of a terrorist organization’s strength, instructions for how to carry out acts of violence, and anything that glorifies the violence itself. Social media is an important part of terrorists’ recruitment strategy, say backers of the legislation. “Whether it was the Nice attacks, whether it was the Bataclan attack in Paris, whether it’s Manchester, [...] they have all had a direct link to online extremist content,” says Lucinda Creighton, a senior adviser at the Counter Extremism Project (CEP), a campaign group that has helped shape the legislation. The new laws require platforms to take down any terrorism-related content within an hour of a notice being issued, force them to use a filter to ensure it’s not reuploaded, and, if they fail in either of these duties, allow governments to fine companies up to 4 percent of their global annual revenue. For a company like Facebook, which earned close to $17 billion in revenue last year, that could mean fines of as much as $680 million (around €600 million). Advocates of the legislation say it’s a set of common-sense proposals that are designed to prevent online extremist content from turning into real-world attacks. But critics, including Internet Freedom think tanks and big tech firms, claim the legislation threatens the principles of a free and open internet, and it may jeopardize the work being done by anti-terrorist groups. The proposals are currently working their way through the committees of the European Parliament, so a lot could change before the legislation becomes law. Both sides want to find a balance between allowing freedom of expression and stopping the spread of extremist content online, but they have very different ideas about where this balance lies. Why is the legislation needed? Terrorists use social media to promote themselves, just like big brands do. Organizations such as ISIS use online platforms to radicalize people across the globe. Those people may then travel to join the organization’s ranks in person or commit terrorist attacks in support of ISIS in their home countries. At its peak, ISIS has had a devastatingly effective social media strategy, which both instilled fear in its enemies and recruited new supporters. In 2019, the organization’s physical presence in the Middle East has been all but eliminated, but the legislation’s supporters argue that this means there’s an even greater need for tougher online rules. As the group’s physical power has diminished, the online war of ideas is more important than ever. The recent attack in New Zealand where the alleged shooter identified as a 28 year old Australian man, Brenton Tarrant announced the attack on the anonymous-troll message board 8chan. He posted images of the weapons days before the attack, and made an announcement an hour before the shooting. On 8chan, Facebook and Twitter, he also posted links to a 74-page manifesto, titled “The Great Replacement,” blaming immigration for the displacement of whites in Oceania and elsewhere. The manifesto cites “white genocide” as a motive for the attack, and calls for “a future for white children” as its goal. Further he live-streamed the attacks on Facebook, YouTube; and posted a link to the stream on 8chan. “Every attack over the last 18 months or two years or so has got an online dimension. Either inciting or in some cases instructing, providing instruction, or glorifying,” Julian King, a British diplomat and European commissioner for the Security Union, told The Guardian when the laws were first proposed. With the increasing frequency with which terrorists become “self-radicalized” by online material shows the importance of the proposed laws. One-hour takedown limit; upload filters & private Terms of Service The one-hour takedown is one of two core obligations for tech firms proposed by the legislation. Under the proposals, each EU member state will designate a so-called “competent authority.” It’s up to each member state to decide exactly how this body operates, but the legislation says they’re responsible for flagging problematic content. This includes videos and images that incite terrorism, that provide instructions for how to carry out an attack, or that otherwise promote involvement with a terrorist group. Once content has been identified, this authority would then send out a removal order to the platform that’s hosting it, which can then delete it or disable access for any users inside the EU. Either way, action needs to be taken within one hour of a notice being issued. It’s a tight time limit, but removing content this quickly is important to stop its spread, according to Creighton. This obligation is similar to voluntary rules that are already in place that encourage tech firms to take down content flagged by law enforcement and other trusted agencies in an hour. Another part is the addition of a legally mandated upload filter, which would hypothetically stop the same pieces of extremist content from being continuously reuploaded after being flagged and removed — although these filters have sometimes been easy to bypass in the past. “The frustrating thing is that [extremist content] has been flagged with the tech companies, it’s been taken down and it’s reappearing a day or two or a week later,” Creighton says, “That has to stop and that’s what this legislation targets.” The other part is the prohibition of content using private Terms of Service (TOS), rather than national law, and to take down more material than the law actually requires. This effectively increases the power of authorities in any EU Member State to suppress information that is legal elsewhere in the EU. For example, authorities in Hungary and authorities in Sweden may disagree about a news organization sharing an interview with a current or former member of a terrorist organization that it is “promoting” or “glorifying” terrorism. Or they may differ on the legitimacy of a civil society organizations advocacy on complex issues in Chechnya, Israel, or Kurdistan. This regulation gives platforms reason to use their TOS to accommodate whichever authority wants such content taken down – and to apply that decision to users everywhere. What’s the problem with the legislation? Critics say that the upload filter could be used by governments to censor their citizens, and that aggressively removing extremist content could prevent non-governmental organizations from being able to document events in war-torn parts of the world. One prominent opponent is the Center for Democracy and Technology (CDT), a think tank funded in part by Amazon, Apple, Facebook, Google, and Microsoft. Earlier this year, it published an open letter to the European Parliament, saying the legislation would “drive internet platforms to adopt untested and poorly understood technologies to restrict online expression.” The letter was co-signed by 41 campaigners and organizations, including the Electronic Frontier Foundation, Digital Rights Watch, and Open Rights Group. “These filtering technologies are certainly being used by the big platforms, but we don’t think it’s right for government to force companies to install technology in this way,” the CDT’s director for European affairs, Jens-Henrik Jeppesen, told The Verge in an interview. Removing certain content, even if a human moderator has correctly identified it as extremist in nature, could prove disastrous for the human rights groups that rely on them to document attacks. For instance, in the case of Syria’s civil war, footage of the conflict is one of the only ways to prove when human rights violations occur. But between 2012 and 2018, Google took down over 100,000 videos of attacks that were carried out in Syria’s civil war, which destroyed vital evidence of what took place. The Syrian Archive, an organization that aims to verify and preserve footage of the conflict, has been forced to backup footage on its own site to prevent the records from disappearing. Opponents of the legislation like the CDT also say that the filters could end up acting like YouTube’s frequently criticized Content ID system. This ID allows copyright owners to file takedowns on videos that use their material, but the system will sometimes remove videos posted by their original owners, and they can misidentify original clips as being copyrighted. It can also be easily circumvented. Opponents of the legislation also believe that the current voluntary measures are enough to stop the flow of terrorist content online. They claim the majority of terrorist content has already been removed from the major social networks, and that a user would have to go out of their way to find the content on a smaller site. “It is disproportionate to have new legislation to see if you can sanitize the remaining 5 percent of available platforms,” Jeppesen says. These organizations need to be able to view this content, no matter how troubling it might be, in order to investigate war crimes. Their independence from governments is what makes their work valuable, but it could also mean they’re shut out under the new legislation. While Lucinda doesn’t believe free and public access to this information is the answer. She argues that needing to “analyze and document recruitment to ISIS in East London” isn’t a good enough excuse to leave content on the internet if the existence of that content “leads to a terrorist attack in London, or Paris or Dublin.” The legislation is currently working its way through the European Parliament, and its exact wording could yet change. At the time of publication, the legislation’s lead committee is currently due to vote on its report on the draft regulation. After that, it must proceed through the trilogue stage — where the European Commission, the Council of the European Union, and the European Parliament debate the contents of the legislation — before it can finally be voted into law by the European Parliament. Because the bill is so far away from being passed, neither its opponents nor its supporters believe a final vote will take place any sooner than the end of 2019. That’s because the European Parliament’s current term ends next month, and elections must take place before the next term begins in July. Here’s the link to the proposed bill by the European Commission. How social media enabled and amplified the Christchurch terrorist attack Tech regulation to an extent of sentence jail: Australia’s ‘Sharing of Abhorrent Violent Material Bill’ to Warren’s ‘Corporate Executive Accountability Act’ EU’s antitrust Commission sends a “Statements of Objections” to Valve and five other video game publishers for “geo-blocking” purchases

Some time ago I started gathering the SQL Saturday XML files. I started to parse and work with these, but the data is a mess and it never was that important. I regret that now. In any case, with the announcement of insolvency from PASS recently, I asked people to save some data. One of those that responded was Ben Weissman (b|t), and he actually rendered out all the available schedules as PDFs (using VB, of course). He sent them to me and I decided to load them up. Tl;dr; You can see my page here: https://dataplatformdaysweb.azurewebsites.net/ I had started an initial project for replacing SQL Saturday, but hadn’t planned on anything more than a static site. Actually, I wanted a sortable grid, but that was beyond the time and web skills I had at this time. That’s still a goal. This is a quick look at how I built things. I am not a real software developer, at least not on the current, interactive web. This stuff likely makes sense to any junior web dev, but it was learning for me. Azure DevOps and Dot Net Core I wanted a simple site, but I wanted this built in Azure DevOps, so it has some control and tracking. I thought about a simple HTML site, but producing a build with that didn’t seem intuitive to me, so I fired up Visual Studio. I chose a Dot Net Core ASP.NET REACT application, as I may move this to Linux. It’s cheaper In any  case, I took the defaults. No real reason other than I’ve tried MVC and that was hard, and lots of people seem to like react. I also have people I can bug at Redgate. I got the default project to build locally. Then I changed the names of the pages and loaded this into an Azure DevOp repo. Once up there, I took a default build process. I pointed this at my repo and then clicked Save and Queue… and it failed. Changes to the Build I got a message that the nuget restore wouldn’t work with dotnet core 3.1. I could fall back to 2.2, but when I did that, the project wouldn’t build locally. I realized I’d initially selected a Windows VS-2016 hosted agent, but I had built the project on VS2019. I changed that to the Windows 2019 agent and it worked. Deployment to Azure I’d set up an Azure App Service already, and I created a very simple release step. I linked my artifact and selected a release to an Azure App Service Plan. I had to authorize my plan, but once I did that, I was able to select the App Service I’d set up. No configuration needed. I clicked save, built a release, and ran. I found the default React Site at my URL. Changes to the Project I made a few changes to the project as well, to remove some of the defaults. First, I needed to load my PDFs into the project. I had originally created an Assets folder in the root of the project, but that did not get included in the artifact that was built. Looking at the project, and searching around Google a bit, led me to see that the main page, index.html, was in the ClientApp/public folder. I moved my Assets folder below this, and then saw all my files included in the build artifact and deployed. I also wanted to remove some of the default sample menu items. I found these in the ClientApp/src/components folder in the NavMenu.js. I deleted the two entries, leaving just a “home” there for now. I may do some other grouping later. Building the Archive This was the more interesting item for me. Ben had sent me a ZIP file with all the PDF files in it. I unzipped these and I saw this view: Originally I thought a simple list of numbers and files would get me started, but there are hundreds of files. How can I do this? My first thought as PowerShell can help. I popped this open and use Get-ChildItem to get a list of files and compile this into a variable. I have been wanting to use Azure Data Studio more for PoSh, and that’s where I did this. This got me a basic HTML list of my files. I had trouble with the pathing, so rather than futz around and try to build production code here, I just used this and then a “search and replace” of the [a href=”] to add a [a href=”/Assets/PDF/”] got me the correct paths. I need to learn how to properly get paths working here in PoSh, but this string manipulation wasn’t important for a one off task. Once I had this, I had something. Of course, at this point, Ben sent me his index list of the event names, which was what I really wanted. I could have taken the source of his page and used search and replace to get the pathing, but I did something stupider. In a hurry, I copied and pasted his list of events into SSMS in a new Query Window. One of the reasons I do this is that the cross line editing is superior (IMHO) to VS and VSCode. I’ll repeat the process with just a few lines here, but keep in mind I had like 800. This is a useful text trick as well for some data changing. I had this list: I wanted to make this a table, so I use the Select+Alt+Arrows to select the entire first column. I then added my table HTML. I could do this in VSCode, but the reason I like SSMS is that I can space over to the right and then get a vertical edit link, rather than a set of end-of-line cursors. I then can create another edit point and add other code, like this: I wrapped this in the table beginning and ending and had my table. What about the URLS? Well, I could easily add the paths, but then getting the individual file names was hard. Or was it? I used the same trick. I pasted my list code into SSMS and selected all the file names: I copied and pasted this set of vertical text into my table, and viola, I had a working table that looked, well, about as good as I could make it quickly. More to come, as I try to archive and preserve the SQL Saturday data and history as best I can. The post Building a SQL Saturday Archive appeared first on SQLServerCentral.

Time series are a very common data format that describes how things change over time. Some of the most common sources are industrial machines and IoT devices, IT infrastructure stacks (such as hardware, software, and networking components), and applications that share their results over time. Managing time series data efficiently is not easy because the data model doesn’t fit general-purpose databases. For this reason, I am happy to share that Amazon Timestream is now generally available. Timestream is a fast, scalable, and serverless time series database service that makes it easy to collect, store, and process trillions of time series events per day up to 1,000 times faster and at as little as to 1/10th the cost of a relational database. This is made possible by the way Timestream is managing data: recent data is kept in memory and historical data is moved to cost-optimized storage based on a retention policy you define. All data is always automatically replicated across multiple availability zones (AZ) in the same AWS region. New data is written to the memory store, where data is replicated across three AZs before returning success of the operation. Data replication is quorum based such that the loss of nodes, or an entire AZ, does not disrupt durability or availability. In addition, data in the memory store is continuously backed up to Amazon Simple Storage Service (S3) as an extra precaution. Queries automatically access and combine recent and historical data across tiers without the need to specify the storage location, and support time series-specific functionalities to help you identify trends and patterns in data in near real time. There are no upfront costs, you pay only for the data you write, store, or query. Based on the load, Timestream automatically scales up or down to adjust capacity, without the need to manage the underlying infrastructure. Timestream integrates with popular services for data collection, visualization, and machine learning, making it easy to use with existing and new applications. For example, you can ingest data directly from AWS IoT Core, Amazon Kinesis Data Analytics for Apache Flink, and Amazon MSK. You can visualize data stored in Timestream from Amazon QuickSight, and use Amazon SageMaker to apply machine learning algorithms to time series data, for example for anomaly detection. You can use Timestream fine-grained AWS Identity and Access Management (IAM) permissions to easily ingest or query data from an AWS Lambda function. We are providing the tools to use Timestream with open source platforms such as Apache Kafka, Telegraf, Prometheus, and Grafana. Using Amazon Timestream from the Console In the Timestream console, I select Create database. I can choose to create a Standard database or a Sample database populated with sample data. I proceed with a standard database and I name it MyDatabase. All Timestream data is encrypted by default. I use the default master key, but you can use a customer managed key that you created using AWS Key Management Service (KMS). In that way, you can control the rotation of the master key, and who has permissions to use or manage it. I complete the creation of the database. Now my database is empty. I select Create table and name it MyTable. Each table has its own data retention policy. First data is ingested in the memory store, where it can be stored from a minimum of one hour to a maximum of a year. After that, it is automatically moved to the magnetic store, where it can be kept up from a minimum of one day to a maximum of 200 years, after which it is deleted. In my case, I select 1 hour of memory store retention and 5 years of magnetic store retention. When writing data in Timestream, you cannot insert data that is older than the retention period of the memory store. For example, in my case I will not be able to insert records older than 1 hour. Similarly, you cannot insert data with a future timestamp. I complete the creation of the table. As you noticed, I was not asked for a data schema. Timestream will automatically infer that as data is ingested. Now, let’s put some data in the table! Loading Data in Amazon Timestream Each record in a Timestream table is a single data point in the time series and contains: The measure name, type, and value. Each record can contain a single measure, but different measure names and types can be stored in the same table. The timestamp of when the measure was collected, with nanosecond granularity. Zero or more dimensions that describe the measure and can be used to filter or aggregate data. Records in a table can have different dimensions. For example, let’s build a simple monitoring application collecting CPU, memory, swap, and disk usage from a server. Each server is identified by a hostname and has a location expressed as a country and a city. In this case, the dimensions would be the same for all records: country city hostname Records in the table are going to measure different things. The measure names I use are: cpu_utilization memory_utilization swap_utilization disk_utilization Measure type is DOUBLE for all of them. For the monitoring application, I am using Python. To collect monitoring information I use the psutil module that I can install with: pip3 install psutil Here’s the code for the collect.py application: import time import boto3 import psutil from botocore.config import Config DATABASE_NAME = "MyDatabase" TABLE_NAME = "MyTable" COUNTRY = "UK" CITY = "London" HOSTNAME = "MyHostname" # You can make it dynamic using socket.gethostname() INTERVAL = 1 # Seconds def prepare_record(measure_name, measure_value): record = { 'Time': str(current_time), 'Dimensions': dimensions, 'MeasureName': measure_name, 'MeasureValue': str(measure_value), 'MeasureValueType': 'DOUBLE' } return record def write_records(records): try: result = write_client.write_records(DatabaseName=DATABASE_NAME, TableName=TABLE_NAME, Records=records, CommonAttributes={}) status = result['ResponseMetadata']['HTTPStatusCode'] print("Processed %d records. WriteRecords Status: %s" % (len(records), status)) except Exception as err: print("Error:", err) if __name__ == '__main__': session = boto3.Session() write_client = session.client('timestream-write', config=Config( read_timeout=20, max_pool_connections=5000, retries={'max_attempts': 10})) query_client = session.client('timestream-query') dimensions = [ {'Name': 'country', 'Value': COUNTRY}, {'Name': 'city', 'Value': CITY}, {'Name': 'hostname', 'Value': HOSTNAME}, ] records = [] while True: current_time = int(time.time() * 1000) cpu_utilization = psutil.cpu_percent() memory_utilization = psutil.virtual_memory().percent swap_utilization = psutil.swap_memory().percent disk_utilization = psutil.disk_usage('/').percent records.append(prepare_record('cpu_utilization', cpu_utilization)) records.append(prepare_record( 'memory_utilization', memory_utilization)) records.append(prepare_record('swap_utilization', swap_utilization)) records.append(prepare_record('disk_utilization', disk_utilization)) print("records {} - cpu {} - memory {} - swap {} - disk {}".format( len(records), cpu_utilization, memory_utilization, swap_utilization, disk_utilization)) if len(records) == 100: write_records(records) records = [] time.sleep(INTERVAL) I start the collect.py application. Every 100 records, data is written in the MyData table: $ python3 collect.py records 4 - cpu 31.6 - memory 65.3 - swap 73.8 - disk 5.7 records 8 - cpu 18.3 - memory 64.9 - swap 73.8 - disk 5.7 records 12 - cpu 15.1 - memory 64.8 - swap 73.8 - disk 5.7 . . . records 96 - cpu 44.1 - memory 64.2 - swap 73.8 - disk 5.7 records 100 - cpu 46.8 - memory 64.1 - swap 73.8 - disk 5.7 Processed 100 records. WriteRecords Status: 200 records 4 - cpu 36.3 - memory 64.1 - swap 73.8 - disk 5.7 records 8 - cpu 31.7 - memory 64.1 - swap 73.8 - disk 5.7 records 12 - cpu 38.8 - memory 64.1 - swap 73.8 - disk 5.7 . . . Now, in the Timestream console, I see the schema of the MyData table, automatically updated based on the data ingested: Note that, since all measures in the table are of type DOUBLE, the measure_value::double column contains the value for all of them. If the measures were of different types (for example, INT or BIGINT) I would have more columns (such as measure_value::int and measure_value::bigint) . In the console, I can also see a recap of which kind measures I have in the table, their corresponding data type, and the dimensions used for that specific measure: Querying Data from the Console I can query time series data using SQL. The memory store is optimized for fast point-in-time queries, while the magnetic store is optimized for fast analytical queries. However, queries automatically process data on all stores (memory and magnetic) without having to specify the data location in the query. I am running queries straight from the console, but I can also use JDBC connectivity to access the query engine. I start with a basic query to see the most recent records in the table: SELECT * FROM MyDatabase.MyTable ORDER BY time DESC LIMIT 8 Let’s try something a little more complex. I want to see the average CPU utilization aggregated by hostname in 5 minutes intervals for the last two hours. I filter records based on the content of measure_name. I use the function bin() to round time to a multiple of an interval size, and the function ago() to compare timestamps: SELECT hostname, bin(time, 5m) as binned_time, avg(measure_value::double) as avg_cpu_utilization FROM MyDatabase.MyTable WHERE measure_name = 'cpu_utilization' AND time > ago(2h) GROUP BY hostname, bin(time, 5m) When collecting time series data you may miss some values. This is quite common especially for distributed architectures and IoT devices. Timestream has some interesting functions that you can use to fill in the missing values, for example using linear interpolation, or based on the last observation carried forward. More generally, Timestream offers many functions that help you to use mathematical expressions, manipulate strings, arrays, and date/time values, use regular expressions, and work with aggregations/windows. To experience what you can do with Timestream, you can create a sample database and add the two IoT and DevOps datasets that we provide. Then, in the console query interface, look at the sample queries to get a glimpse of some of the more advanced functionalities: Using Amazon Timestream with Grafana One of the most interesting aspects of Timestream is the integration with many platforms. For example, you can visualize your time series data and create alerts using Grafana 7.1 or higher. The Timestream plugin is part of the open source edition of Grafana. I add a new GrafanaDemo table to my database, and use another sample application to continuously ingest data. The application simulates performance data collected from a microservice architecture running on thousands of hosts. I install Grafana on an Amazon Elastic Compute Cloud (EC2) instance and add the Timestream plugin using the Grafana CLI. $ grafana-cli plugins install grafana-timestream-datasource I use SSH Port Forwarding to access the Grafana console from my laptop: $ ssh -L 3000:<EC2-Public-DNS>:3000 -N -f ec2-user@<EC2-Public-DNS> In the Grafana console, I configure the plugin with the right AWS credentials, and the Timestream database and table. Now, I can select the sample dashboard, distributed as part of the Timestream plugin, using data from the GrafanaDemo table where performance data is continuously collected: Available Now Amazon Timestream is available today in US East (N. Virginia), Europe (Ireland), US West (Oregon), and US East (Ohio). You can use Timestream with the console, the AWS Command Line Interface (CLI), AWS SDKs, and AWS CloudFormation. With Timestream, you pay based on the number of writes, the data scanned by the queries, and the storage used. For more information, please see the pricing page. You can find more sample applications in this repo. To learn more, please see the documentation. It’s never been easier to work with time series, including data ingestion, retention, access, and storage tiering. Let me know what you are going to build! — Danilo

Image retrieval and image recognition are fundamental problems in the machine learning and computer vision world. Image classification technology has shown remarkable progress over the past few years. An obstacle in this research, however, is the unavailability of large annotated datasets. Google has made an attempt to solve this challenge by introducing Google-Landmarks, a worldwide dataset for recognition of human-made and natural landmarks. This dataset was made with the intention of solving fine-grained and instance-level recognition problems. Examples of this include identifying important landmarks in images (Eiffel Tower, Mount Fuji, Taj Mahal, etc), which accounts for a large portion of what people like to photograph. Landmark recognition can help predict landmark labels directly from image pixels to help people better understand and organize their photo collections. The Google-Landmarks dataset contains more than 2 million images depicting 30 thousand unique landmarks from across the world, a number of classes that is almost 30x larger than what is available in commonly used datasets. Geographic distribution of landmarks in the Landmark dataset Google has also open-sourced Deep Local Features DELF, an attentive local feature descriptor, which is useful for large-scale instance-level image recognition, in order to advance research in this area. DELF detects and describes semantic local features which can be geometrically verified between images showing the same object instance. It is also optimized for landmark recognition. Google-Landmarks is being released as part of the Landmark Recognition and Landmark Retrieval Kaggle challenges. The Landmark recognition challenge calls for developers to build models that recognize the correct landmark (if any) in a dataset of challenging test images. In the retrieval challenge, developers are given query images and for each query, they are expected to retrieve all database images containing the same landmarks (if any). Participants are encouraged to compete in both these challenges as the test set for both the problems is same. Participants may also use the training data from the recognition challenge to train models which could be useful for the retrieval challenge. However, there are no landmarks in common between the training/index sets of the two challenges. This challenge is the focal point of the CVPR’18 Landmarks workshop. More details of the challenge and the dataset can be found in the Google research blog.  

A group of researchers from Stanford University and University of California with iRhythm Technologies Inc. and Veterans Affairs Palo Alto Health Care System have build a model that can help in the diagnosis of irregular heart rhythms, also called as arrhythmias. On Monday, the researchers shared their findings in a paper published on Springer Nature: Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Detecting arrhythmias is a pretty easy task for an expert technician or a cardiologist but is known to be quite challenging for computers. With the help of widely available ECG data and deep learning, this study aimed to improve the accuracy and scalability of automated ECG analysis. For this study, the researchers built a 34-layer deep neural network (DNN) and trained it to detect arrhythmia in arbitrary length ECG time series. The model was trained on 91,232 single-lead ECGs from 53,549 patients who used a single-lead ambulatory ECG monitoring device. The network learned to classify noise and the sinus rhythm. Additionally, it also learned to classify and segment twelve arrhythmia types present in the time series. For testing the model, the researchers used an independent test dataset annotated by a consensus committee of board-certified practicing cardiologists. The test dataset used in this study is publicly available at iRhythm Technologies’s GitHub repository. The model did pretty well by achieving an average area under the receiver operating characteristic curve (ROC) of 0.97. Another measure of accuracy was F1, which is a harmonic mean of the positive predictive value and sensitivity. F1 score of the DNN (0.837) exceeded that of average cardiologists (0.780). Researchers introduce a CNN-based system for identifying radioresistant cancer-causing cells Stanford researchers introduce DeepSolar, a deep learning framework that mapped every solar panel in the US Our healthcare data is not private anymore: Study reveals that machine learning can be used to re-identify individuals from physical activity data

Last week, Facebook, in a Securities and Exchange Commission filing, announced its annual stockholders' meeting, which will be held on May 30. Going by the proposals listed in the notice looks like the main agenda of this meeting is to make changes to Facebook’s governance structure. The notice lists eight stockholders proposals that Facebook investors will be casting their votes for. One of these proposals is titled as “a stockholder proposal regarding an independent chair,” indicating that the investors will be deciding on whether Mark Zuckerberg should step down as the chairman and be replaced by an independent hire. In recent years, we have witnessed an endless number of scandals in which Facebook played a major part, the main one being the Cambridge Analytica data scandal. These scandals have left the investors, who boast of having nearly $3 million shares of the company, in anger and frustration. They believe that Facebook has been unable to properly address these issues because its current corporate governance structure gives Zuckerberg dual power of both CEO and chairman. Facebook has a dual-class structure, Class A and B. Class B shares have 10 times the voting power of class A shares and not-so-surprisingly Zuckerberg has more than 75% of class B stock. This makes him the holder of more than half of the voting power at Facebook and therefore allows him to dismiss the investor proposals. Last year in July, Trillium Asset Management, which manages about $11 million (£8.4 million) in Facebook stock, wrote a proposal to oust Mark Zuckerberg as chairman of the social-networking giant. "A CEO who also serves as a chair can exert excessive influence on the board and its agenda, weakening the board's oversight of management. Separating the chair and CEO positions reduces this conflict, and an independent chair provides the clearest separation of power between the CEO and the rest of the board," reads the proposal by Trillium Asset Management. Facebook has previously said that splitting Zuckerberg’s role would create "uncertainty, confusion, and inefficiency in board and management function." This time, however, Facebook’s board of directors have called for voting against the stockholders’ proposal. Facebook added, “Our board of directors currently believes that the most effective leadership model is that Mr. Zuckerberg, our founder, and controlling stockholder, serves as both Chairman and CEO.” Their response further says, “We believe our board of directors is functioning effectively under its current structure, and that the current structure provides appropriate oversight protections. We do not believe that requiring the Chair to be independent will provide appreciably better direction and performance, and instead could cause inefficiency in board and management function and relations.” Looking at the previous responses by Facebook on such proposals, chances of Zuckerberg stepping down as chair is very similar. “They have zero chance of succeeding. There's no way in hell that Zuckerberg will voluntarily relinquish his position at the top or his control of so much voting power,” adds a Redditor. Another user adds,  “Won't matter. Zuckerberg has special shares that give him essentially majority votes by himself”. Read the full story at Business Insider. Facebook AI introduces Aroma, a new code recommendation tool for developers Facebook AI open-sources PyTorch-BigGraph for faster embeddings in large graphs Ahead of Indian elections, Facebook removes hundreds of assets spreading fake news and hate speech, but are they too late?

Microsoft announced an open-source model for Component Firmware Update (CFU), for Windows developers. CFU enables delivering firmware updates for peripheral components through Windows Update by using CFU drivers. This protocol aims to enable system and peripheral developers to leverage the CFU protocol. It allows them to easily and automatically push firmware updates to Windows Update for their firmware components. CFU aims to bring smooth updates via Windows updates and verify the firmware version before download. CFU permits but does not specify authentication, encryption, rollback policies/ methods, or recovery of bricked firmware. Overview of CFU The CFU driver is the host and is created by the device manufacturer. It is delivered via a Windows Update. Then the driver is installed once the device is detected by Windows. Primary and sub-components A hierarchical system with a primary component and subcomponents is followed in a CFU compatible system. A primary component implements CFU on the device side and can receive updates for itself and the connected sub-components. A device may have multiple primary components with or without additional sub-components. Offers and payloads A CFU driver which is the host, may contain multiple firmware images for a primary component and its sub-components. A package in the host consists of an offer, a payload and other information. The offer contains information about the payload to allow the primary component in deciding if it is acceptable. A payload is the firmware image. Offer Sequence The primary component can accept, reject, or skip the offer of firmware update. On accepting, the payload is delivered immediately. On rejecting or skipping, the host cycles through all other offers in the list. Host independence The host’s (CFU driver) decisions are independent of the offers’ contents or payloads. It does not necessarily use any logic and simply sends the offers and the accepted payloads. Payload delivery On an offer being accepted, the host proceeds to download the firmware image or referred as the payload. Delivery is done in three phases—beginning, middle, and end. The payload is a set of addresses and fixed-size arrays of bytes. Payload validation and authentication Validation of the incoming firmware update is an important aspect. The primary component should verify bytes after each write ensuring that the data is stored properly before proceeding with the next set of data bytes. A CRC or hash should also be calculated on download, to be verified after the download is complete, ensuring the data wasn’t modified in transit. In addition, a cryptographic signature mechanism is recommended to provide end-to-end protection. An encryption mechanism can also be employed for confidential downloads. On image authentication, the properties should be validated against the offer and other rules the device manufacturer may specify. CFU does not specify any rules to be applied. Payload Invocation The CFU Protocol is run at the application level in the primary component. The component can continue to do other tasks as long it can receive and store the incoming payload without significant disruptions. The only real disruption occurs when the new firmware must be invoked. There are two recommended ways to avoid that disruption. A very generic approach is to use a small bootloader image to select one of multiple images to run when the device is reset. This is typically at boot time. The image selection algorithm is specific to the implementation. It is typically based on an algorithm which involves code version, and an indication of successful image validation. Another invocation method is to physically swap the memory of the desired image with the active address space. This is done upon reset. A disadvantage of this method is that it requires specialized hardware. The advantage being all images are statically linked to the same address space eliminating the need for a bootloader. CFU limitations There are some limitations of the protocol. It cannot update a bricked component that can no longer run the protocol. CFU does not provide any security. The CFU protocol requires extra memory to store the incoming images which helps in non-disruptive updates. Updating sub-component images larger than the component’s available storage requires dividing the sub-component image into smaller packages The CFU protocol allows pausing the download, so care needs to be taken for proper validation. CFU assumes that the primary component has set validation rules. If they need to be changed, the component must first be successfully updated by using the old rules first, only then new rules can be applied. For more details, visit the Microsoft website. How the Titan M chip will improve Android security Microsoft fixing and testing the Windows 10 October update after file deletion bug Paul Allen, Microsoft co-founder, philanthropist, and developer dies of cancer at 65

RedisGraph was released in beta mode six months ago. On the 14th of November, RedisLabs announced the general availability of RedisGraph v1.0. RedisGraph is a Redis module that adds a graph database functionality to Redis. RedisGraph delivers a fast and efficient way to store, manage and process graphs, around 6 to 600 times faster than existing graph databases. RedisGraph represents connected data as adjacency matrices and employs the power of GraphBLAS which is a highly optimized library for sparse matrix operations. How does RedisGraph Work? Redis is a single-threaded process by default. RedisGraph is bound to the single thread of Redis and supports all incoming queries while including a threadpool that takes a configurable number of threads at the module’s loading time to handle higher throughputs. The queries are calculated in one of the threads of the threadpool. This means reads can scale and handle large throughput easily. Each query only runs in one thread. This is what separates RedisGraph from other graph database implementations—which typically execute each query on all available cores of the machine. This makes RedisGraph more suitable for real-time and real-world use cases where high throughput and low latency under concurrent operations are important. In RedisGraph, a write query ( that modifies the graph in any way ) must be executed in complete isolation. RedisGraph also ensures write/readers separation by using a read/write (R/W) lock. This means that either multiple readers can acquire the lock or just a single writer can write a query.  The lock ensures that as long as a writer is executing, no one can acquire the lock, and as long as there’s a reader executing, no writer can obtain the lock. Benchmarking RedisGraph The team conducted a benchmark test on RedisGraph that proved the latter’s speed was more than other graph databases. They used a simple benchmark released by TigerGraph that covered the following: Data loading time Storage size of loaded data Query response time for k-hop neighborhood count Query response time for weakly connected components and page rank The TigerGraph benchmark compared all other graph databases and reported TigerGraph to be 2-8000 times faster than any other graph database. The Redis team compared RedisGraph using the exact same setup. The test focused mainly on the k-hop neighborhood count query. To test the result of concurrent operations, parallel requests were added to the TigerGraph benchmark. RedisGraph utilized just a single core and other graph databases were using up to 32 cores. ReddisGraph was faster in response times than any other graph database (with the exception of TigerGraph in the single request k-hop queries tests on the Twitter dataset). The single request benchmark test and parallel request benchmark test also returned positive results for RedisGraph. In all the tests conducted, RedisGraph never timed out or generated out of memory exceptions. RedisGraph shows performance improvements under load of 6 to 60 times faster than existing graph solutions for a large dataset (twitter dataset) and 20 to 65 times faster on a normal data set (graph500 dataset). The benchmark also proves that RedisGraph outperforms Neo4j, Neptune, JanusGraph and ArangoDB on a single request response time with improvements almost 36 to 15,000 times faster. There were 2X and 0.8X faster single request response times as compared to TigerGraph. Future improvements as listed by the team include: Performance improvements for aggregations and large result sets A faster version of GraphBLAS More Cypher clauses/functionality to support even more diverse queries Integration for graph visualization software LDBC benchmarking tests You can head over to RedisLabs official Blog to know more about the benchmarking tests conducted. Introducing Web High Level Shading Language (WHLSL): A graphics shading language for WebGPU Facebook’s GraphQL moved to a new GraphQL Foundation, backed by The Linux Foundation 2018 is the year of graph databases. Here’s why.

After the successful conference at Amsterdam, Vuejs recently conducted the first ever VueConf.US in New Orleans on March 26th-28th 2018. The conference congregated hundreds of Vuejs developers from around the world and the VueJS Core team and featured workshops and talks from members of the Vuejs community and Vue experts. It also witnessed new releases and project processes. The conference commenced with an inaugural keynote by Evan You, the creator of Vue. He talked about the growth of Vue since 2016, also highlighting new developments coming soon. In his keynote, You said that “Vue will be moving to a standardized release cycle with new minor releases every three months and a minimum six-month notice prior to major releases.” They will also be shifting from a single release channel to four separated release channels. The VueConf.US conference covered 4 major workshops by Vuejs experts. In the first workshop, Evan You talked about building simple versions of libraries for features such as routing, state management, form validation and i18n using basic Vue features. Chris Fritz, conducted a second workshop on the basics of building world-class Vue applications. Topics included configuring Webpack for single-file components, setting up the most advanced workflows currently possible, and more. Sarah Drasner organized a workshop on animation in Vue to creating complex effects in performant and visually stunning patterns. Rachel Nabors, presented a talk, "Vue In Motion", on implementing animations and transitions in Vue. In the fourth workshop, Blake Newman presented his views on Vuex, a state management pattern. The VueConf.US conference also featured multiple presentations by key Vue experts. Daniel Rosenwasser, Program Manager on TypeScript at Microsoft, presented his views on making TypeScript and Vue seamless to make sure that JavaScript users of all communities can use Typescript. In another interesting presentation, Jen Looper talked about creating an Engaging Native Mobile App with Vue and NativeScript all the while retaining shared code between the Vue created website and mobile app. Edd Yerburgh, Vue core team member and author of "vue-test-utils", spoke about testing Vue applications. He presented an adapted the testing pyramid for the front end, consisting of end-to-end tests, snapshot tests, and unit tests. In a talk, “Vue & SSR: The best practices”, Sebastien Chopin talks about common problems with server-side rendering and how to deal with them. He also shows Nuxt.js as a possible solution to most of these problems. Community support was the highlight of the VueConf.US conference with a large number of talks mentioning the importance of Vue community support. The attendees were extremely positive about the conference and stated how comfortable and welcoming the community was. Each attendee shared a similar level of excitement about the platform and its possibilities and was excited about meeting the wide variety of developers and to know their experiences. All talks were recorded and will be posted soon on VueMastery.

Yesterday, at the MariaDB OpenWorks 2019, the annual user and developer conference of the database, the MariaDB Corporation team announced a new, fully open source MariaDB Enterprise Server. The MariaDB Enterprise Server will support customers by delivering a database engineered for greater reliability and stability. It will be used as the default version for customers for on-premise or in the cloud. Max Mether, VP of Server Product Management at MariaDB Corporation said to us in an email, "We're seeing that our enterprise customers have very different needs from the average community user. These customers are working on a completely different scale with a strong focus on stability and security. In order to be able to cater to these requirements, it is clear that we need to focus on a different solution by creating another version of MariaDB Server specifically focused on enterprise production workloads." Features of MariaDB Enterprise Server New enterprise-centric features MariaDB Enterprise Server comes with features that solve specific enterprise requirements. The new features in development include enhanced MariaDB Backup, improved audit plugin, and full data-at-rest encryption of MariaDB Cluster. Security, performance and scalability for production MariaDB Enterprise Server is configured for secure, high-performance production environments, unlike Community Server. It provides reliable and faster backups for large databases. It provides end-to-end encryption for all data at rest in MariaDB clusters. Stability at scale The MariaDB Enterprise Server goes through rigorous quality assurance and testing and is pre-configured to fulfill the requirements of secure production environments. Release Integrity MariaDB Enterprise Server is distributed securely with a clearly established chain of custody from MariaDB to customers for ensuring that binaries cannot be tampered with. Pat Casey, SVP of Development and Operations, ServiceNow, said to us via an email, "Thousands of the world's largest organizations depend on the Now Platform to create great experiences and unlock productivity. Better quality assurance and stability of critical enterprise features are extremely compelling. At our scale and in production with 100,000 MariaDB databases, reliability is what matters most." MariaDB Enterprise Server 10.4 is available with next version of MariaDB Platform in spring 2019. The team will also release GA versions of MariaDB Enterprise Server 10.2 and 10.3, this spring, that will include high-end enterprise features, such as enhanced Backup. Amazon’s Mark Porter joins MariaDB as the advisor to board of directors The company further announced that Mark Porter, who ran the Amazon Relational Database Service (RDS) recently, joined MariaDB as an advisor to the board of directors. Porter is currently the CTO at Grab, a transportation and mobile payments company. He has previously served as the vice president at Oracle Corporation. Porter said, "MariaDB's DBaaS solutions give businesses many advantages. By focusing on customer needs and using their deep database expertise, they have built optimizations, flexibility and enterprise capabilities that no one else can deliver. With MariaDB's growing popularity as an option to escape Oracle, the opportunity is extremely strong to capture large market share and delight customers. I'm both humbled and thrilled to be part of the MariaDB team as relational databases continue to run the most important companies on the internet." According to the team at MariaDB, Mark Porter will contribute his expertise of cloud, distributed systems and database operations to help MariaDB rapidly grow its database-as-a-service (DBaaS) offering. He will also work in the direction of growth for SkySQL, and will further integrate new distributed technology into MariaDB Platform. Michael Howard, CEO, MariaDB Corporation, said, "Mark's guidance will be a tremendous asset in building a next-generation MariaDB cloud. Mark has a proven record of operating and scaling database services while driving rapid growth. SkySQL is designed from the ground up to offer the best MariaDB service for multi-cloud, including private cloud environments. It offers enterprise product capabilities beyond the MariaDB community server, that is used widely in public clouds, to ensure quality of service, security and features otherwise only found in proprietary legacy databases." TiDB open sources its MySQL/MariaDB compatible data migration (DM) tool MariaDB acquires Clustrix to give database customers ‘freedom from Oracle lock-in’ MariaDB 10.3.7 releases

Yesterday, Google announced Flutter Release Preview 2, during the keynote of Google Developer Days in Shanghai. This is the final preview before the Google team releases Flutter 1.0. In this preview release, they have expanded support for the "Cupertino" themed controls and executing Dart code in the background and reduced the package size. Flutter is Google’s new open-source mobile app SDK using which you can quickly create high-quality native interfaces on iOS and Android. What’s new in Flutter Release Preview 2? Extended support for Cupertino themed controls After getting the feedback on Flutter Release Preview 1, this release is designed with keeping Apple interface guidelines in mind. They have expanded support for the "Cupertino" themed controls in Flutter, with an extensive library of widgets and classes. Some of the added iOS-themed widgets are: CupertinoApp, a convenience widget that wraps a number of widgets that are commonly required for an iOS-design targeting application. CupertinoTimerPicker is used to show countdown duration with hour, minute and second spinners. CupertinoSegmentedControl displays the widgets provided in the Map of children in a horizontal list. It is used to select between a number of mutually exclusive options. CupertinoActionSheet is used for a specific style of alert that presents the user with a set of two or more choices related to the current context. Support for executing Dart code in the background In this release, support has been added for executing Dart code in the background, even while the application is suspended.   Reduced package size The application package size is now reduced by up to 30% on both Android and iOS. A minimal Flutter app on Android now weighs just 4.7 MB when built in release mode, and they are continually working towards identifying further potential optimizations. How to upgrade to Flutter Release Preview 2? If you're using the beta release already, you can upgrade to Flutter Release Preview 2 just by running the following: $ flutter upgrade Follow the instructions mentioned on the Flutter blog for upgrading to Flutter Release Preview 2. To know more about Flutter Preview 2 in detail, check out the official announcement by Google. Google Flutter moves out of beta with release preview 1 Google’s Dart hits version 2.0 with major changes for developers Is Google planning to replace Android with Project Fuchsia?

After Google just got saved from GDPR’s huge fine last month, Twitter is next on the EU’s GDPR investigation checklist. This appears to be the first GDPR investigation to be opened against Twitter. Last week, the data privacy regulators in Ireland opened up an investigation against Twitter’s data collection practices. This is to analyze the amount of data Twitter receives from its URL-shortening system, t.co. Twitter says the URL shortening allows the platform to measure the number of clicks per link, and helps it to fight the spread of malware through suspicious links. Why did Irish data regulators choose to investigate Twitter? This news was first reported by Fortune stating, “Michael Veale, who works at University College London, suspects that Twitter gets more information when people click on t.co links, and that it might use them to track those people as they surf the web, by leaving cookies in their browsers.” Veale asked Twitter to provide him with all the personal data it holds on him. To which, Twitter refused claiming that providing this information would take a disproportionate effort. Following this, Veale filed a complaint to the Irish Data Protection Commission (DPC), and the authorities opened an investigation last week. In a letter to Veale, the Irish Data Privacy Commissioner wrote, “The DPC has initiated a formal statutory inquiry in respect of your complaint. The inquiry will examine whether or not Twitter has discharged its obligations in connection with the subject matter of your complaint and determine whether or not any provisions of the GDPR or the [Irish Data Protection] Act have been contravened by Twitter in this respect.” The Irish authorities said that Veale’s complaint will be handled by the new European Data Protection Board as Veale’s complaint involves cross-border processing. The EU Data protection body helps national data protection authorities coordinate their GDPR enforcement efforts. Veale also prompted a similar investigation probe into Facebook, which also refused to hand over data held on users’ web-browsing activities. However, Fortune says, “ Facebook was already the subject of multiple GDPR investigations.” Veale says, "Data which looks a bit creepy, generally data which looks like web-browsing history, [is something] companies are very keen to keep out of data access requests. The user has a right to understand." Twitter, however, refused to comment, saying only that it was ‘actively engaged’ with the DPC. If Twitter is found to be in GDPR’s breach list, it could face a fine of up to €20m or up to 4 percent of its global annual revenue. To know more about this news in detail, head over to Fortune’s full coverage. How Twitter is defending against the Silhouette attack that discovers user identity GDPR is good for everyone: businesses, developers, customers The much loved reverse chronological Twitter timeline is back as Twitter attempts to break the ‘filter bubble’

After the successful launch of Elon Musk’s mammoth space mission, Starlink last week, the company has unveiled a brand new website with more details on the Starlink commercial satellite internet service. Starlink Starlink sent 60 communications satellites to the orbit which will eventually be part of a single constellation providing high speed internet to the globe. SpaceX has plans to deploy nearly 12,000 satellites in three orbital shells by the mid-2020s, initially placing approximately 1600 in a 550-kilometer (340 mi)-altitude area. The new website gives a few glimpses of how Starlink’s plan looks like such as including the CG representation of how the satellites will work. These satellites will move along their orbits simultaneously, providing internet in a given area. They have also revealed more intricacies about the satellites. Flat Panel Antennas In each satellite, the signal is transmitted and received by four high-throughput phased array radio antennas. These antennas have a flat panel design and can transmit in multiple directions and frequencies. Starlink Ion Propulsion system and solar array Each satellite carries a krypton ion propulsion system. These systems enable satellites to orbit raise, maneuver in space, and deorbit. There is also a singular solar array, singe for simplifying the system. Ion thrusters provide a more fuel-efficient form of propulsion than conventional liquid propellants. It uses Krypton, which is less expensive than xenon but offers lower thrust efficiency. Starlink Star Tracker and Autonomous collision avoidance system Star Tracker is Space X’s inbuilt sensors, that can tell each satellite’s output for precise broadband throughput placement and tracking. The collision avoidance system uses inputs from the U.S. Department of Defense debris tracking system, reducing human error with a more reliable approach. Through this data it can perform maneuvers to avoid collision with space debris and other spacecrafts. Per Techcrunch, who interviewed a SpaceX representative, “the debris tracker hooks into the Air Force’s Combined Space Operations Center, where trajectories of all known space debris are tracked. These trajectories are checked against those of the satellites, and if a possible collision is detected the course changes are made, well ahead of time.” Source: Techcrunch More information on Starlink (such as the cost of the project, what ground stations look like, etc) is yet unknown. Till that time, keep an eye on the Starlink’s website and this space for new updates. SpaceX delays launch of Starlink, its commercial satellite internet service, for the second time to “update satellite software” Jeff Bezos unveils space mission: Blue Origin’s Lunar lander to colonize the moon Elon Musk reveals big plans with Neuralink

In a paper published earlier this month, a team of AI researchers at Facebook have been looking closely at how AI agents ‘understand’ images and the extent to which they can be said to develop a shared conceptual language. Building on earlier research that indicates “(AI) agents are now developing conceptual symbol meanings,” the Facebook research team attempted to dive deeper and look closely at how AI agents develop representations of visual inputs. What they found was intriguing - the conceptual ‘language’ that the AI agents seemed to share wasn’t in any way related to the input data, but instead what the researchers describe as a ‘shared conceptual pact’. This research is significant as it opens the lid on how agents in deep learning systems, and opens up new possibilities for understanding how they work. Background Researchers take their cue from current research into AI agents. This research runs visual ‘games’..“This… allows us to address the exciting issue of whether the needs of goal-directed communication will lead agents to associate visually-grounded conceptual representations to discrete symbols, developing natural language-like word meanings” reads the paper. However, most of the existing studies present only the analysis of the agents’ symbol usage. Very little attention is given to the representation of the visual input developed by the agents during the interaction process.  Researchers have made use of the referential games of Angeliki Lazaridou, a research scientist at Deepmind, where a pair of agents communicates about images using a fixed-size vocabulary. “Unlike in those previous studies, which suggested that the agents developed a shared understanding of what the images represented, our researchers found that they extracted no concept-level information”, reads the research paper. The paired AI agents would arrive at an image-based decision depending only on the low-level feature similarities. How does it work? Researchers implemented Lazaridou’s, same-image game and the different image game. In the same image game, the Sender and Receiver are shown the same two images (that are always of different concepts). In the different-image game, the Receiver is shown different images than the Sender’s every time. The experiments were repeated using 100 random initialization seeds. Researchers first looked at how playing the game affects the way agents “see” the input data. This involves figuring out which of the image embeddings differ from the input image representations, and from each other. Researchers then further predicted that as the training continues, Sender and Receiver representations become quite similar to each other, as well as the input ones. To finally compare the similarity structure of the input, Sender and the Receiver spaces, representational similarity analysis (RSA) from computational neuroscience is used by the researchers. AI agents reach an image-based consensus The paired agents in the game arrived at an ‘image-based consensus’ depending solely on low-level feature similarities, without determining, for instance, that pictures of a Boston terrier and a Chihuahua both represent dogs. In fact, the agents were able to reach this consensus despite being presented with similar patterns of visual noise, which included no recognizable objects. This confirmed the hypothesis that the Sender and Receiver are capable of communicating about the input data with no conceptual content at all. This, in turn, suggests that no concept-level information (e.g., features that would allow to identify the instances of the dog or chair category) has been extracted by the agents during the training process. For more information, check out the official research paper. UK researchers have developed a new PyTorch framework for preserving privacy in deep learning Researchers show that randomly initialized gradient descent can achieve zero training loss in deep learning UK researchers build the world’s first quantum compass to overthrow GPS

Yesterday Apple talked about exclusivity enforcement in Swift 5, in a post. No this is not some exclusive feature or patenting of some sort. This idea is on how variables in a Swift program access memory. Swift is the programming language used for developing Apple apps. What is exclusivity enforcement? The Swift 5 release allows runtime checks on “Exclusive Access to Memory”. This further adds to Swift showing that it is a ‘safe language’. For memory safety to take place, Swift needs exclusive access to a variable and modify it. This means that the variable can be accessed only with the same name when it is being modified as particular arguments. A programmer’s intention in case of exclusivity violations is often ambiguous in Swift. So, to protect against it and to allow the safety features, exclusivity enforcement was introduced in Swift 4. In Swift 4, both compile-time and run-time enforcement was available, the latter being available only in debug builds. Some of the holes in exclusivity enforcement are patched in Swift 5 by changing the language model. So runtime exclusivity enforcement is enabled by default in Release builds. This can impact Swift projects in two ways: Violation of Swift exclusivity rules causing a runtime trap Overhead due to memory access checks can degrade performance slightly Why exclusivity enforcement? This enforcement is done mainly to enforce memory safety in Swift. It eliminates dangerous interactions in Swift programs which involves mutable states Enforcement gets rid of unspecified behavior rules from Swift It is mandatory to maintain ABI stability In addition to protecting memory safety, this enforcement helps in optimizing performance The exclusivity rules give programmers the control to move only types Even though the memory problem is a rare occurrence, addressing it early on improves Swift a bit. A comment on Hacker news says: “The benefit being that you only have to deal with this issue rarely, rather than all the time with manual memory management.” Apple is patenting Swift features like optional chaining Swift 5 for Xcode 10.2 beta is here with stable ABI Swift is now available on Fedora 28