Machine learners
His seminal publication “Neural Networks and Physical Systems with Emergent Collective Computational Abilities“ opened up the field by formulating memory storage as a Lyapunov function. He later turned to working on the representation of time in olfactory cortex.
Who is John Hopfield?
This model of a biological neuron won the creators a Nobel prize for its elegant description of the activation and inactivation of ionic currents in a neuron.
What is a Hodgkin Huxley model?
This network beat the world champion Go player in 2017
What is AlphaGo?
It is the name of the rule originally described in 1949 as “When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A’s efficiency, as one of the cells firing B, is increased.”
What is the Hebbian learning rule?
It is currently the most popular GPU manufacturer for deep learning.
What is NVIDIA?
He reportedly said that “it took 17 years to get deep learning right; one year thinking and 16 years of progress in computing, praise be to Intel.”
Who is Geoff Hinton?
This cortical cell preferentially responds to oriented edges.
What is a simple cell?
This pioneering CNN was created to demonstrate how handwritten characters could be recognized reliably after being trained on tens of thousands of examples.
What is LeNet?
This algorithm for improving DNN training was created by Srivastava and Hinton and others, inspired by observations of neural pruning during early brain development.
What is dropout?
This spiking neural network chip was developed by a DARPA and successfully demonstrated a number of small but very fast deep neural network applications despite having only binary synapses; however, its extremely large chip size and a fixed amount of memory limits mass production.
What is TrueNorth?
He died in a mysterious sailboat incident exactly on his 43rd birthday, in part due to the criticism by MIT’s Marvin Minsky of his seminal work on single layer networks, which led to a long stall in work on artificial neural networks.
Who is Frank Rosenblatt?
Some of these visual cells differentially respond to stimuli that are closer to or further away than the point of fixation.
What are binocular cells?
This network which was devised by Teuvo Kohonen about 1990 self-constructs to cluster data into connected low dimensional map regions.
What is a self organizing (feature) map?
It is the acronym for a neural network hardware accelerator in a smartphone application processors, analogous to GPU for graphics processing unit.
What is an NPU?
He was a protege of Jean Piaget, developed the Logo programming language, and co-authored one of the most influential early works on machine learning.
Who was Seymour Papert?
This cortical area has many cells responsive to complete objects, in particular, to human and animal faces.
What is area IT?
This pioneering hardware network implementation of brain computing was created in the 1980s by Kunihiko Fukushima and was used to demonstrate handwritten character recognition about 15 years before the creation of the MNIST dataset.
What is the Neocognitron?
This is a model-free, off-policy type of reinforcement learning that computes action-value functions which indicate the expected utility of taking a certain action in a given state, while following the optimal policy afterwards.
What is Q learning?
1nJ is the approximate energy cost for retrieving one 32-bit word of data from this type of memory.
What is DRAM?
His 1988 publication of “Learning to Predict by the Methods of Temporal Differences” underlies all of modern reward learning.
Who is Richard Sutton?
The number 86 billion is often quoted.
What are the number of neurons in the human brain?
This network developed in the late 1980s was one of the first to establish methods for temporal credit assignment, in work from Gerry Tesauro, on a classic dice game of luck and skill; two keys to its success were the clever representation of game states and the use of temporal difference learning.
What is TD-Gammon?
This error backpropagation training method randomly but gradually reduce the mistakes that a network makes and is the dominant training method used nowadays.
What is stochastic gradient descent?
This massive publicly funded project used mixed signal technology to build wafer scale neural processors that work “faster than real time”, but was slow to produce convincing applications.
What is BrainScales?