Neural networks
Pre-deep-learning era.
26 milestones used this technique.
UMass Amherst researchers demonstrate protein-nanowire memristors for neuromorphic computing
In June 2021, researchers at the University of Massachusetts Amherst published findings in Nature Communications showing that protein nanowires harvested from the bacterium Geobacter sulfurreducens can function as memristors, enabling brain-inspired computation at ultralow power without a conventional battery.
Swinburne University of Technology Researchers Demonstrate Optical Neuromorphic Processor Using Micro-Comb Photonic Chip
In January 2021, a Swinburne University of Technology-led international team published results in Nature demonstrating an optical neuromorphic processor built on a photonic micro-comb chip, achieving a processing speed of 10 TOPS from a single integrated device and performing image classification tasks at high speed.
Analogue resistive memory circuit solves linear algebra problems in one step
In March 2019, Daniele Ielmini and colleagues at Politecnico di Milano published results in PNAS demonstrating a crosspoint resistive-memory circuit that solves linear systems, matrix eigenvector problems, and differential equations by physical analogue relaxation, substantially reducing the energy and latency costs of conventional iterative digital solvers.
Blue Brain Project Digital Reconstruction of the Rat Somatosensory Cortex Microcircuitry
On 8 October 2015, Henry Markram and colleagues at the Blue Brain Project published a detailed computational reconstruction of 31,000 neurons and 37 million synapses in a 0.29 mm³ column of juvenile rat somatosensory cortex, creating the first large-scale digital model of a mammalian cortical microcircuit.
Neurorobotics Platform, Human Brain Project
In 2015, the Human Brain Project, a European Commission Flagship Initiative, released its Neurorobotics Platform, a simulation environment allowing researchers to connect large-scale brain models to virtual robot bodies and run closed-loop cognitive experiments without physical hardware.
Neural Turing Machine Introduced by Alex Graves, Greg Wayne, and Ivo Danihelka
In October 2014, Alex Graves, Greg Wayne, and Ivo Danihelka at Google DeepMind published 'Neural Turing Machines', a preprint proposing a neural network architecture augmented with an external memory matrix and differentiable read/write operations, enabling the system to learn algorithms such as sorting and copying from examples alone.
IBM TrueNorth: a neuromorphic integrated circuit with one million programmable neurons
In August 2014, researchers at IBM Research published a description of TrueNorth, a neuromorphic chip containing one million programmable spiking neurons and 256 million synapses on a 4096-core CMOS integrated circuit, built under the DARPA SyNAPSE programme.
Spaun: A Large-Scale Functional Brain Model Capable of Performing Multiple Cognitive Tasks
In November 2012, Chris Eliasmith and colleagues at the University of Waterloo published a description of Spaun (Semantic Pointer Architecture Unified Network) in Science, presenting a 2.5-million-neuron brain model capable of performing eight distinct cognitive tasks, including digit recognition, question answering, and list recall.
Google Brain Unsupervised Neural Network Learns to Detect Cats from YouTube Frames
In June 2012, Quoc V. Le and colleagues at Google Brain published research showing that a 1,000-machine, 16,000-core neural network trained without labels on 10 million YouTube thumbnail images spontaneously developed a neuron selectively responsive to human and cat faces, demonstrating large-scale unsupervised feature learning from unlabelled video data.
BrainGate2 Participants Use Thought-Controlled Robotic Arm to Reach and Grasp
In May 2012, researchers in the BrainGate2 clinical trial, led by Leigh Hochberg and colleagues at Massachusetts General Hospital, Brown University, and affiliated institutions, demonstrated that two participants with tetraplegia could use neural signals decoded from a 96-electrode intracortical array to control a robotic arm and perform reach-and-grasp tasks without manual assistance.
IBM Simulates 4.5 Percent of Human Brain Activity Using Blue Gene Supercomputer
In November 2011, IBM researchers led by Dharmendra Modha at IBM Research Almaden demonstrated a cortical simulation on the Blue Gene/P supercomputer that modelled approximately 4.5 percent of human-scale neural activity, using 147,456 processors to represent 1.617 billion neurons and 8.87 trillion synapses.
Google Brain Founded by Andrew Ng and Jeff Dean
In 2011, Andrew Ng and Jeff Dean co-founded Google Brain, an internal research group at Google dedicated to large-scale deep learning. The project demonstrated that deep neural networks trained on substantial compute could learn useful representations without labelled data, reshaping how the industry approached machine learning research.
Nengo Neural Simulation Software Released by the Computational Neuroscience Research Group, University of Waterloo
Researchers at the Computational Neuroscience Research Group (CNRG) at the University of Waterloo developed Nengo, an open-source software environment for simulating large-scale neural systems using the Neural Engineering Framework, providing tools that bridge high-level network specification with low-level neurophysiological detail.
Silicon Retina with Ganglion Cell Spiking Outputs as Neural Prosthesis
Around 2006, researchers developed a silicon retina implemented as an analogue VLSI chip that modelled four primary retinal ganglion cell types and generated 3,600 spiking outputs, designed as a neural prosthesis matched to the physical dimensions of the biological retina.
Numenta Founded by Jeff Hawkins and Donna Dubinsky
In 2005, Jeff Hawkins and Donna Dubinsky co-founded Numenta, a research company dedicated to developing machine intelligence systems modelled on the structural and algorithmic principles of the mammalian neocortex, building on Hawkins's theoretical framework published in his 2004 book On Intelligence.
iCub Humanoid Robot Platform
The iCub humanoid robot was developed from 2004 onward by the RobotCub Consortium, coordinated by the Istituto Italiano di Tecnologia (IIT), as an open-hardware child-sized platform for research into embodied cognition, motor learning, and human–robot interaction.
Fujitsu Laboratories Develops Dynamically Reconfigurable Neural Network for Humanoid Robot Motor Learning
In March 2003, Fujitsu Laboratories announced a dynamically reconfigurable neural network system enabling humanoid robots to learn motor coordination from experience, reducing learning time that previously required days or months and substantially cutting the volume of motion-control software needed.
A Neural Probabilistic Language Model by Yoshua Bengio and Colleagues
In 2003, Yoshua Bengio, Réjean Ducharme, Pascal Vincent, and Christian Janvin at the Université de Montréal published 'A Neural Probabilistic Language Model' in JMLR, demonstrating that a feed-forward neural network trained on word sequences could learn distributed word representations and outperform n-gram models on perplexity benchmarks.
Torch Machine Learning Library
In 2002, Ronan Collobert, Samy Bengio, and Johnny Mariéthoz at IDIAP Research Institute published a paper introducing Torch, a modular C++ and Lua-scriptable machine learning library that unified a range of algorithms, including support vector machines and neural networks, under a common object-oriented framework.
MIT Cog Project: Humanoid Robotics for Cognitive Research
Around 1993, Rodney Brooks and colleagues at the Massachusetts Institute of Technology Artificial Intelligence Laboratory initiated the Cog project, constructing an upper-torso humanoid robot intended to investigate whether human-like cognitive capacities could emerge from embodied interaction with a physical environment.
TD-Gammon Developed by Gerald Tesauro at IBM
In 1992, Gerald Tesauro at IBM Thomas J. Watson Research Center developed TD-Gammon, a backgammon program that trained itself through self-play using temporal-difference learning applied to a multilayer neural network, reaching a standard of play close to that of strong human experts.
Yann LeCun Applies Backpropagation to Handwritten ZIP Code Recognition at AT&T Bell Labs
In 1989, Yann LeCun and colleagues at AT&T Bell Labs published 'Backpropagation Applied to Handwritten Zip Code Recognition', demonstrating that a convolutional neural network trained with backpropagation could read handwritten postal ZIP codes with high accuracy, establishing a template for practical deep learning in computer vision.
ALVINN: An Autonomous Land Vehicle in a Neural Network, Developed by Dean Pomerleau at CMU
In 1989, Dean Pomerleau at Carnegie Mellon University published ALVINN (Autonomous Land Vehicle in a Neural Network), a three-layer backpropagation network trained on road images that steered the CMU Navlab vehicle autonomously, demonstrating that a neural network could learn driving behaviour directly from sensor data.
NETtalk Neural Network Developed by Terrence J. Sejnowski and Charles Rosenberg
Terrence J. Sejnowski of the Salk Institute and Charles Rosenberg of Princeton University developed NETtalk, a feedforward neural network trained to convert English text to speech, publishing the principal account in Complex Systems in 1987. The network learned pronunciation from examples alone, demonstrating that a multi-layer perceptron could acquire a complex linguistic skill without hand-coded rules.
Backpropagation Described by Arthur E. Bryson Jr. and Yu-Chi Ho
In 1969, Arthur E. Bryson Jr. and Yu-Chi Ho of Harvard University described a gradient-based optimisation procedure for multi-stage dynamic systems in their textbook Applied Optimal Control, presenting what is now recognised as an early statement of the backpropagation principle in a supervised-learning context.
Perceptrons: An Introduction to Computational Geometry
In 1969, Marvin Minsky and Seymour Papert of MIT published Perceptrons: An Introduction to Computational Geometry, a formal mathematical analysis of single-layer perceptrons that demonstrated key limitations, notably the inability to compute non-linearly separable functions such as XOR, and contributed to a reduction in funding and research activity in connectionist approaches to AI.