GRAPH NAME ARTIFICIAL INTELLIGENCE AND ML *** ## NODE 1 NAME MCCULLOCH-PITTS NEURON DATE 1943 PLACE USA WHO Warren McCulloch and Walter Pitts BRIEF DESCRIPTION McCulloch and Pitts propose a simplified mathematical model of neurons that combines binary inputs through logical rules. The work establishes an early connection between neural networks, logic, and computation and constitutes a formal precursor to connectionism. LINK *** ## NODE 2 NAME TURING TEST DATE 1950 PLACE United Kingdom WHO Alan Turing BRIEF DESCRIPTION Alan Turing publishes Computing Machinery and Intelligence and proposes the imitation game as an operational criterion for discussing intelligent behavior in machines. The proposal shifts part of the debate from an abstract definition of intelligence toward observable behavior. LINK *** ## NODE 3 NAME DARTMOUTH DATE 1956 PLACE USA - New Hampshire - Hanover WHO John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon BRIEF DESCRIPTION The Dartmouth Summer Research Project on Artificial Intelligence brings researchers together around problems of reasoning, learning, and the automation of cognitive capabilities. The project consolidates artificial intelligence as the name of a distinct scientific field. LINK *** ## NODE 4 NAME LOGIC THEORIST DATE 1956 PLACE USA WHO Allen Newell, Herbert Simon and Cliff Shaw BRIEF DESCRIPTION Logic Theorist proves logic theorems through symbolic search and heuristics. The system constitutes one of the earliest operational demonstrations of automated problem solving through explicit representation of states and rules. LINK *** ## NODE 5 NAME PERCEPTRON DATE 1957 PLACE USA - New York WHO Frank Rosenblatt BRIEF DESCRIPTION Frank Rosenblatt develops the perceptron as a supervised learning model for linear classification. The algorithm automatically adjusts weights from examples and becomes one of the earliest widely studied models of neural learning. LINK *** ## NODE 6 NAME LISP FOR AI DATE 1958 PLACE USA - Massachusetts - Cambridge WHO John McCarthy BRIEF DESCRIPTION LISP introduces a representation based on lists and functions suitable for symbolic manipulation. Its expressiveness for recursion, search, and structure transformation makes it one of the principal languages of artificial intelligence research for decades. LINK *** ## NODE 7 NAME ADALINE DATE 1960 PLACE USA - California - Stanford WHO Bernard Widrow and Marcian Hoff BRIEF DESCRIPTION ADALINE uses an adaptive neural unit whose weights are adjusted through an error-minimization rule. The system constitutes an important precursor to optimization methods for neural models and adaptive signal processing. LINK *** ## NODE 8 NAME DENDRAL DATE 1965 PLACE USA - California - Stanford WHO Edward Feigenbaum, Bruce Buchanan and Joshua Lederberg BRIEF DESCRIPTION DENDRAL applies specialized knowledge rules to infer molecular structures from chemical data. The project demonstrates the value of incorporating domain-specific knowledge and becomes a direct precursor to expert systems. LINK *** ## NODE 9 NAME ELIZA DATE 1966 PLACE USA - Massachusetts - Cambridge WHO Joseph Weizenbaum BRIEF DESCRIPTION ELIZA implements textual interaction through patterns and linguistic transformation rules. Although it did not perform general semantic understanding, the system showed that relatively simple rules could produce conversations that some users interpreted as meaningful. LINK *** ## NODE 10 NAME SHAKEY DATE 1966 PLACE USA - California WHO SRI INTERNATIONAL BRIEF DESCRIPTION Shakey integrates perception, planning, and execution in a computer-controlled mobile robot. The project constitutes one of the earliest systems to combine environmental representation, symbolic reasoning, and physical actions within an autonomous architecture. LINK *** ## NODE 11 NAME MYCIN DATE 1972 PLACE USA - California - Stanford WHO STANFORD UNIVERSITY BRIEF DESCRIPTION MYCIN uses production rules and certainty factors to assist reasoning about bacterial infections and treatments. It becomes a representative example of the expert-system approach based on explicitly encoded knowledge. LINK *** ## NODE 12 NAME FIRST AI WINTER DATE 1974 PLACE WHO ARTIFICIAL INTELLIGENCE COMMUNITY BRIEF DESCRIPTION During the 1970s, funding and expectations for several artificial intelligence programs decline because of technical limitations, insufficient computing capacity, and results below initial expectations. This period becomes known as the first AI winter. LINK *** ## NODE 13 NAME BAYESIAN NETWORKS DATE 1985 PLACE USA - California WHO Judea Pearl BRIEF DESCRIPTION Bayesian networks consolidate a graphical representation of random variables and conditional dependencies through directed graphs. The approach provides a formal foundation for probabilistic inference and reasoning under uncertainty in intelligent systems. LINK *** ## NODE 14 NAME BACKPROPAGATION DATE 1986 PLACE USA WHO David Rumelhart, Geoffrey Hinton and Ronald Williams BRIEF DESCRIPTION The widespread adoption of the backpropagation algorithm demonstrates a practical procedure for adjusting multilayer neural networks through gradient propagation. The method enables the learning of internal representations and renews interest in neural models with hidden layers. LINK *** ## NODE 15 NAME TD LEARNING DATE 1988 PLACE USA WHO Richard Sutton BRIEF DESCRIPTION Temporal-difference learning combines elements of Monte Carlo methods and dynamic programming to update estimates from successive predictions. The approach becomes a central component of modern reinforcement learning. LINK *** ## NODE 16 NAME Q-LEARNING DATE 1989 PLACE United Kingdom WHO Christopher Watkins BRIEF DESCRIPTION Q-learning introduces a reinforcement learning algorithm capable of estimating action values without requiring an explicit model of environmental transitions. Its update rule enables the approximation of optimal policies through interaction with the environment. LINK *** ## NODE 17 NAME SUPPORT VECTOR MACHINES DATE 1995 PLACE USA WHO Corinna Cortes and Vladimir Vapnik BRIEF DESCRIPTION Support vector machines formalize classification through maximum-margin hyperplanes and kernel functions. The approach provides effective methods for supervised learning and contributes to the development of statistical learning theory. LINK *** ## NODE 18 NAME DEEP BLUE DATE 1997-05-11 PLACE USA - New York - New York WHO IBM BRIEF DESCRIPTION Deep Blue defeats Garry Kasparov in an official six-game chess match. The system combines intensive search, specialized evaluation, and dedicated hardware, demonstrating the potential of specialized computational methods for complex decision problems. LINK *** ## NODE 19 NAME LSTM DATE 1997 PLACE Germany WHO Sepp Hochreiter and Jürgen Schmidhuber BRIEF DESCRIPTION Long Short-Term Memory introduces a recurrent architecture with memory mechanisms designed to preserve information across long sequences. LSTM reduces problems associated with vanishing gradients and later becomes an important architecture for sequential data. LINK *** ## NODE 20 NAME DEEP BELIEF NETWORKS DATE 2006 PLACE Canada - Ontario - Toronto WHO Geoffrey Hinton, Simon Osindero and Yee-Whye Teh BRIEF DESCRIPTION A layer-by-layer training procedure for deep networks using generative models is presented. The work helps renew interest in deep architectures and precedes the later expansion of deep learning through large datasets and computational acceleration. LINK *** ## NODE 21 NAME IMAGENET DATE 2009 PLACE USA - New Jersey - Princeton WHO Fei-Fei Li and ImageNet team BRIEF DESCRIPTION ImageNet establishes a large-scale visual dataset organized into labeled categories. Its availability enables recognition systems to be evaluated on millions of images and makes it an important experimental infrastructure for computer vision. LINK *** ## NODE 22 NAME ALEXNET DATE 2012 PLACE Canada - Ontario - Toronto WHO Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton BRIEF DESCRIPTION AlexNet achieves a substantial improvement in the ImageNet competition using a deep convolutional neural network trained with GPUs. The result accelerates the adoption of deep learning for vision and demonstrates the combined importance of data, models, and computational capacity. LINK *** ## NODE 23 NAME WORD2VEC DATE 2013 PLACE USA WHO GOOGLE BRIEF DESCRIPTION Word2Vec introduces efficient methods for learning vector representations of words from large corpora. Geometric relationships between vectors capture statistical and semantic regularities useful for natural language processing. LINK *** ## NODE 24 NAME GENERATIVE ADVERSARIAL NETWORKS DATE 2014 PLACE Canada WHO Ian Goodfellow and collaborators BRIEF DESCRIPTION Generative adversarial networks introduce competitive training between a generator and a discriminator. This framework enables models to learn complex distributions and becomes an important family of generative models. LINK *** ## NODE 25 NAME ALPHAGO DATE 2016-03-15 PLACE South Korea - Seoul WHO DEEPMIND BRIEF DESCRIPTION AlphaGo completes a 4-1 victory over Lee Sedol using deep neural networks, tree search, and reinforcement learning. The system demonstrates the effective combination of statistical learning and planning in a domain with enormous combinatorial complexity. LINK *** ## NODE 26 NAME TRANSFORMER DATE 2017 PLACE USA WHO GOOGLE BRIEF DESCRIPTION The Transformer architecture uses attention mechanisms as its primary component for modeling dependencies between elements of a sequence. Its design enables more effective parallelization of training than recurrent architectures and becomes the foundation of numerous modern language models. LINK *** ## NODE 27 NAME BERT DATE 2018 PLACE USA WHO GOOGLE BRIEF DESCRIPTION BERT applies bidirectional pretrained transformers to large corpora using self-supervised objectives. The model demonstrates that a general linguistic representation can subsequently be adapted to multiple tasks through fine-tuning. LINK *** ## NODE 28 NAME GPT-3 DATE 2020-05 PLACE USA - California - San Francisco WHO OPENAI BRIEF DESCRIPTION GPT-3 scales an autoregressive transformer-based model to 175 billion parameters and demonstrates learning capabilities through examples provided directly in context. The work advances the study of large-scale foundation models. LINK *** ## NODE 29 NAME CHATGPT DATE 2022-11-30 PLACE USA - California - San Francisco WHO OPENAI BRIEF DESCRIPTION ChatGPT is released as a conversational system based on pretrained generative models adapted for natural-language interaction. Its adoption significantly expands public use of large-scale language models. LINK