As mentioned, the LSTM architecture relies on a series of gates that can independently influence the activation values (a(t-1)), as well as the memory (c(t-1)), from previous timesteps as information flows through an LSTM unit. These values are transformed as the unit spits out the activations (at) and memory (ct) vectors pertaining to the current timestep at each iteration. While their earlier counterparts enter the unit separately, they are allowed to interact with each other in two broad manners. In the following diagram, the gates (denoted with the capital Greek letter gama, or Γ) represent sigmoid activation functions applied to the dot product of their respectively initialized weight matrix, with previous activations and current input:
- Tech Categories
- Best Sellers
- New Releases
- Books
- Videos
- Audiobooks
Tech Categories Popular Audiobooks
- Articles
- Newsletters
- Free Learning
You're reading from Hands-On Neural Networks with Keras
Niloy Purkait is a technology and strategy consultant by profession. He currently resides in the Netherlands, where he offers his consulting services to local and international companies alike. He specializes in integrated solutions involving artificial intelligence, and takes pride in navigating his clients through dynamic and disruptive business environments. He has a masters in Strategic Management from Tilburg University, and a full specialization in data science from Michigan University. He has advanced industry grade certifications from IBM, in subjects like signal processing, cloud computing, machine and deep learning. He is also perusing advanced academic degrees in several related fields, and is a self-proclaimed lifelong learner.
Read more about Niloy Purkait
Unlock this book and the full library FREE for 7 days
Author (1)
Niloy Purkait is a technology and strategy consultant by profession. He currently resides in the Netherlands, where he offers his consulting services to local and international companies alike. He specializes in integrated solutions involving artificial intelligence, and takes pride in navigating his clients through dynamic and disruptive business environments. He has a masters in Strategic Management from Tilburg University, and a full specialization in data science from Michigan University. He has advanced industry grade certifications from IBM, in subjects like signal processing, cloud computing, machine and deep learning. He is also perusing advanced academic degrees in several related fields, and is a self-proclaimed lifelong learner.
Read more about Niloy Purkait