Shelf

Papers

All 162 seed papers. Seed status means candidacy, not canonical status. Papers with harvested evidence link to their breakdown; the rest are an honestly-declared coverage gap, not a zero.

#PaperYearVenueEvidence
0001A Logical Calculus of the Ideas Immanent in Nervous Activity
First mathematical model of the neuron; the conceptual origin of neural networks.
1943Bulletin of Mathematical Biophysicsscored
0002As We May Think
The memex vision; founding document of augmenting human intellect with machines.
1945The Atlanticno evidence yet
0003A Mathematical Theory of Communication
Created information theory; the quantitative substrate of all machine learning.
1948Bell System Technical Journalscored
0004Computing Machinery and Intelligence
Posed 'can machines think', proposed the imitation game; the field's founding question.
1950Mindscored
0005A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence
Coined 'artificial intelligence' and framed the research programme.
1955Proposal (Dartmouth)no evidence yet
0006The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain
First trainable neural classifier; launched learning machines and their first hype cycle.
1958Psychological Reviewscored
0007Some Studies in Machine Learning Using the Game of Checkers
Coined 'machine learning'; first self-improving game program.
1959IBM Journalscored
0008Programs with Common Sense
The Advice Taker; founding statement of logic-based AI and knowledge representation.
1959Mechanisation of Thought Processesno evidence yet
0009Man-Computer Symbiosis
The augmentation-versus-automation agenda that still structures AI debates.
1960IRE Transactions on Human Factorsscored
0010Steps Toward Artificial Intelligence
Early synthesis of search, learning, and planning as the components of AI.
1961Proceedings of the IREno evidence yet
0011GPS: A Program That Simulates Human Thought
General Problem Solver; means-ends analysis and the symbolic cognition paradigm.
1961Lernende Automaten / RANDscored
0012Fuzzy Sets
Founded fuzzy logic; a major non-probabilistic approach to reasoning under vagueness.
1965Information and Controlscored
0013ELIZA: A Computer Program for the Study of Natural Language Communication
First chatbot; revealed the human tendency to project understanding onto machines.
1966Communications of the ACMno evidence yet
0014Some Philosophical Problems from the Standpoint of Artificial Intelligence
Introduced the frame problem and situation calculus.
1969Machine Intelligence 4scored
0015STRIPS: A New Approach to the Application of Theorem Proving to Problem Solving
The planning formalism that dominated automated planning for decades.
1971Artificial Intelligenceno evidence yet
0016Adaptation in Natural and Artificial Systems (foundational monograph)
Genetic algorithms; evolution as a computational search paradigm.
1975University of Michigan Pressno evidence yet
0017Maximum Likelihood from Incomplete Data via the EM Algorithm
The EM algorithm; workhorse of latent-variable estimation.
1977Journal of the Royal Statistical Society Bscored
0018Minds, Brains, and Programs
The Chinese Room argument; the canonical philosophical attack on strong AI.
1980Behavioral and Brain Sciencesscored
0019Neocognitron: A Self-Organizing Neural Network Model
The architectural ancestor of convolutional networks.
1980Biological Cyberneticsno evidence yet
0020Neural Networks and Physical Systems with Emergent Collective Computational Abilities
Hopfield networks; energy-based associative memory that revived the field (2024 Nobel).
1982PNASno evidence yet
0021Self-Organized Formation of Topologically Correct Feature Maps
Self-organizing maps; unsupervised topology-preserving learning.
1982Biological Cyberneticsscored
0022A Learning Algorithm for Boltzmann Machines
Stochastic networks and unsupervised learning of internal representations.
1985Cognitive Sciencescored
0023A Robust Layered Control System for a Mobile Robot
Subsumption architecture; behaviour-based robotics against the symbolic mainstream.
1986IEEE Journal of Robotics and Automationscored
0024Learning Representations by Back-Propagating Errors
Made backpropagation the standard training algorithm; the engine of all deep learning.
1986Naturescored
0025Fusion, Propagation, and Structuring in Belief Networks
Bayesian networks; principled probabilistic reasoning in AI (Turing Award work).
1986Artificial Intelligencescored
0026Learning to Predict by the Methods of Temporal Differences
TD learning; the core idea of modern RL.
1988Machine Learningscored
0027Multilayer Feedforward Networks Are Universal Approximators
Proved neural nets can approximate any function; the theoretical licence for the field.
1989Neural Networksscored
0028A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition
Made HMMs the standard sequential model for two decades of speech and NLP.
1989Proceedings of the IEEEno evidence yet
0029The Symbol Grounding Problem
Named the problem of how symbols acquire meaning; resurfaces in every LLM debate.
1990Physica Dscored
0030Finding Structure in Time
Simple recurrent networks; sequence learning before LSTMs.
1990Cognitive Sciencescored
0031Intelligence Without Representation
Manifesto for embodied intelligence; the strongest internal critique of GOFAI.
1991Artificial Intelligencescored
0032Q-Learning
Model-free value learning with convergence proof.
1992Machine Learningscored
0033Simple Statistical Gradient-Following Algorithms for Connectionist Reinforcement Learning
REINFORCE; the original policy-gradient method.
1992Machine Learningno evidence yet
0034Causal Diagrams for Empirical Research
The do-calculus; the formal foundation of modern causal inference.
1995Biometrikano evidence yet
0035Support-Vector Networks
SVMs; the dominant classifier of the pre-deep-learning era.
1995Machine Learningscored
0036Temporal Difference Learning and TD-Gammon
Self-play RL reaches expert backgammon; the proof of concept for everything later.
1995Communications of the ACMno evidence yet
0037Regression Shrinkage and Selection via the Lasso
L1 regularization; sparse models across statistics and ML.
1996Journal of the Royal Statistical Society Bscored
0038Long Short-Term Memory
Solved vanishing gradients for sequences; powered a decade of speech and language AI.
1997Neural Computationscored
0039A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting
AdaBoost; proved weak learners can be combined into strong ones.
1997Journal of Computer and System Sciencesscored
0040No Free Lunch Theorems for Optimization
No algorithm wins on all problems; a standing caution against universal claims.
1997IEEE Transactions on Evolutionary Computationscored
0041Gradient-Based Learning Applied to Document Recognition
LeNet; the canonical demonstration that CNNs work on real tasks.
1998Proceedings of the IEEEscored
0042The PageRank Citation Ranking: Bringing Order to the Web
Link-based ranking; the algorithm that organized the web's information.
1998Stanford Technical Reportscored
0043Random Forests
The most-used classical ML algorithm in applied practice.
2001Machine Learningscored
0044Statistical Modeling: The Two Cultures
Named the split between data modeling and algorithmic prediction; prophetic for ML's rise.
2001Statistical Sciencescored
0045Greedy Function Approximation: A Gradient Boosting Machine
Gradient boosting; the backbone of tabular ML to this day.
2001Annals of Statisticsscored
0046BLEU: A Method for Automatic Evaluation of Machine Translation
The metric that made MT progress measurable, virtues and pathologies included.
2002ACLno evidence yet
0047Latent Dirichlet Allocation
Topic models; a decade of probabilistic text analysis.
2003Journal of Machine Learning Researchscored
0048A Neural Probabilistic Language Model
Word embeddings plus neural language modeling; the LLM lineage starts here.
2003Journal of Machine Learning Researchscored
0049A Fast Learning Algorithm for Deep Belief Nets
Layer-wise pretraining; the paper that relaunched 'deep' learning.
2006Neural Computationscored
0050Reducing the Dimensionality of Data with Neural Networks
Deep autoencoders in Science; put deep learning back on the mainstream agenda.
2006Sciencescored
0051Universal Intelligence: A Definition of Machine Intelligence
The formal definition behind much AGI discourse.
2007Minds and Machinesno evidence yet
0052The Basic AI Drives
Instrumental convergence; why capable agents acquire resources and resist shutdown.
2008AGI Conferenceno evidence yet
0053Matrix Factorization Techniques for Recommender Systems
The Netflix-Prize-era standard for collaborative filtering.
2009IEEE Computerscored
0054ImageNet: A Large-Scale Hierarchical Image Database
The dataset that made the deep learning revolution measurable.
2009CVPRscored
0055The Unreasonable Effectiveness of Data
Data beats cleverness; the empirical creed of the scaling era, stated early.
2009IEEE Intelligent Systemsscored
0056ImageNet Classification with Deep Convolutional Neural Networks
AlexNet; the result that started the modern deep learning era.
2012NeurIPSno evidence yet
0057Deep Neural Networks for Acoustic Modeling in Speech Recognition
Deep nets replace decades of speech-recognition engineering.
2012IEEE Signal Processing Magazinescored
0058Fairness Through Awareness
The formal-fairness research programme begins.
2012ITCSscored
0059A Few Useful Things to Know About Machine Learning
The most-shared practical wisdom paper in ML.
2012Communications of the ACMscored
0060Efficient Estimation of Word Representations in Vector Space
word2vec; cheap, composable word meaning ('king - man + woman').
2013ICLR Workshopscored
0061Dropout: A Simple Way to Prevent Neural Networks from Overfitting
The defining regularization trick of the era.
2014Journal of Machine Learning Researchscored
0062Generative Adversarial Networks
GANs; adversarial training created the modern generative-media era.
2014NeurIPSscored
0063Auto-Encoding Variational Bayes
VAEs; probabilistic deep generative modeling.
2014ICLRscored
0064Intriguing Properties of Neural Networks
Discovered adversarial examples; founded ML security.
2014ICLRscored
0065GloVe: Global Vectors for Word Representation
The other standard embedding; count-based meets predictive.
2014EMNLPscored
0066Sequence to Sequence Learning with Neural Networks
Encoder-decoder; end-to-end sequence transduction.
2014NeurIPSscored
0067Very Deep Convolutional Networks for Large-Scale Image Recognition
VGG; depth as the design principle.
2015ICLRno evidence yet
0068Going Deeper with Convolutions
GoogLeNet/Inception; efficient depth via multi-scale modules.
2015CVPRscored
0069Batch Normalization: Accelerating Deep Network Training
Made very deep networks trainable in practice.
2015ICMLno evidence yet
0070Adam: A Method for Stochastic Optimization
The default optimizer of deep learning; among the most-cited papers in CS.
2015ICLRscored
0071Deep Learning (review)
The field's self-definition by its three Turing-Award founders.
2015Natureno evidence yet
0072U-Net: Convolutional Networks for Biomedical Image Segmentation
The default segmentation architecture, far beyond medicine.
2015MICCAIno evidence yet
0073Distilling the Knowledge in a Neural Network
Knowledge distillation; the basis of model compression.
2015NeurIPS Workshopscored
0074Explaining and Harnessing Adversarial Examples
FGSM; explained and weaponized adversarial fragility.
2015ICLRscored
0075Neural Machine Translation by Jointly Learning to Align and Translate
Introduced attention; the mechanism that became everything.
2015ICLRscored
0076Human-Level Control Through Deep Reinforcement Learning
DQN on Atari; deep RL is born.
2015Naturescored
0077Hidden Technical Debt in Machine Learning Systems
Why ML systems rot in production; founding text of MLOps.
2015NeurIPSno evidence yet
0078XGBoost: A Scalable Tree Boosting System
The implementation that made gradient boosting ubiquitous in practice.
2016KDDno evidence yet
0079Deep Residual Learning for Image Recognition
ResNet; skip connections enabled networks of arbitrary depth.
2016CVPRscored
0080You Only Look Once: Unified, Real-Time Object Detection
YOLO; real-time detection as a single network pass.
2016CVPRscored
0081WaveNet: A Generative Model for Raw Audio
Neural audio generation; transformed speech synthesis.
2016arXivno evidence yet
0082Mastering the Game of Go with Deep Neural Networks and Tree Search
AlphaGo; the cultural turning point of the deep learning era.
2016Naturescored
0083End-to-End Training of Deep Visuomotor Policies
Pixels-to-torques; deep learning enters robotic control.
2016Journal of Machine Learning Researchno evidence yet
0084Concrete Problems in AI Safety
Turned AI safety into a concrete ML research agenda.
2016arXivno evidence yet
0085Why Should I Trust You? Explaining the Predictions of Any Classifier
LIME; model-agnostic local explanation.
2016KDDno evidence yet
0086Big Data's Disparate Impact
The canonical legal analysis of algorithmic discrimination.
2016California Law Reviewno evidence yet
0087Machine Bias
The COMPAS investigation; algorithmic injustice becomes front-page news.
2016ProPublicano evidence yet
0088Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
Demonstrated social bias embedded in learned representations.
2016NeurIPSno evidence yet
0089Equality of Opportunity in Supervised Learning
Defined equalized odds; a standard fairness criterion.
2016NeurIPSno evidence yet
0090The Ethics of Algorithms: Mapping the Debate
The standard map of algorithmic-ethics concerns.
2016Big Data & Societyno evidence yet
0091Deep Learning with Differential Privacy
DP-SGD; private training as a practical method.
2016CCSscored
0092Semi-Supervised Classification with Graph Convolutional Networks
GCNs; the breakthrough that mainstreamed graph deep learning.
2017ICLRscored
0093Attention Is All You Need
The Transformer; the architecture of the modern AI era.
2017NeurIPSscored
0094Mastering the Game of Go Without Human Knowledge
AlphaGo Zero; superhuman play from self-play alone.
2017Naturescored
0095Proximal Policy Optimization Algorithms
PPO; the workhorse algorithm, later the engine of RLHF.
2017arXivno evidence yet
0096Deep Reinforcement Learning from Human Preferences
Learning reward from human comparisons; the seed of RLHF.
2017NeurIPSno evidence yet
0097A Unified Approach to Interpreting Model Predictions
SHAP; game-theoretic attribution, the industry default.
2017NeurIPSscored
0098Inherent Trade-Offs in the Fair Determination of Risk Scores
Proved popular fairness definitions are mutually incompatible.
2017ITCSno evidence yet
0099Semantics Derived Automatically from Language Corpora Contain Human-Like Biases
Bias in embeddings, demonstrated with psychometric rigor.
2017Sciencescored
0100Membership Inference Attacks Against Machine Learning Models
Showed models leak whether your data was in the training set.
2017IEEE S&Pno evidence yet
0101Communication-Efficient Learning of Deep Networks from Decentralized Data
Federated learning; training without centralizing data.
2017AISTATSno evidence yet
0102World Models
Agents learning inside their own learned simulators; ancestor of today's world-model agenda.
2018NeurIPSno evidence yet
0103Improving Language Understanding by Generative Pre-Training
GPT-1; generative pretraining as the recipe.
2018OpenAI Technical Reportno evidence yet
0104A General Reinforcement Learning Algorithm That Masters Chess, Shogi, and Go Through Self-Play
AlphaZero; one algorithm, three games.
2018Scienceno evidence yet
0105The Mythos of Model Interpretability
Disciplined the field's vocabulary about what 'interpretable' means.
2018ACM Queue / CACMno evidence yet
0106Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification
Audit that changed commercial face-analysis products and founded audit culture.
2018FAT*no evidence yet
0107Counterfactual Explanations Without Opening the Black Box
Linked explanation methods to GDPR; the legal-technical bridge.
2018Harvard Journal of Law & Technologyno evidence yet
0108The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation
The first broad misuse-threat assessment across digital, physical, political security.
2018arXivno evidence yet
0109BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Pretrain-finetune became the default NLP paradigm.
2019NAACLno evidence yet
0110Language Models are Unsupervised Multitask Learners
GPT-2; scale yields zero-shot task behaviour, and the first staged-release debate.
2019OpenAI Technical Reportno evidence yet
0111Grandmaster Level in StarCraft II Using Multi-Agent Reinforcement Learning
AlphaStar; RL in a real-time, partially observed strategy game.
2019Natureno evidence yet
0112Risks from Learned Optimization in Advanced Machine Learning Systems
Mesa-optimization and deceptive alignment; core inner-alignment concepts.
2019arXivno evidence yet
0113Stop Explaining Black Box Machine Learning Models for High Stakes Decisions
Argues for inherently interpretable models where stakes are high.
2019Nature Machine Intelligenceno evidence yet
0114Model Cards for Model Reporting
The documentation standard for released models.
2019FAT*no evidence yet
0115Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations
Showed a deployed health algorithm systematically disadvantaged Black patients.
2019Scienceno evidence yet
0116Energy and Policy Considerations for Deep Learning in NLP
Put training cost and carbon on the research agenda.
2019ACLno evidence yet
0117The Global Landscape of AI Ethics Guidelines
Mapped 84 guidelines; showed convergence on principles and divergence on practice.
2019Nature Machine Intelligenceno evidence yet
0118The Bitter Lesson
Compute-leveraging general methods beat human-knowledge engineering; the era's most-quoted essay.
2019Essay (incompleteideas.net)no evidence yet
0119On the Measure of Intelligence
Skill-acquisition efficiency as the definition of intelligence; basis of the ARC benchmark.
2019arXivno evidence yet
0120Denoising Diffusion Probabilistic Models
Made diffusion the dominant image-generation paradigm.
2020NeurIPSno evidence yet
0121Neural Radiance Fields (NeRF): Representing Scenes for View Synthesis
Learned 3D scene representation; new field of neural rendering.
2020ECCVno evidence yet
0122Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
T5; everything is text-to-text.
2020Journal of Machine Learning Researchno evidence yet
0123Language Models are Few-Shot Learners
GPT-3; in-context learning and the scaling thesis made undeniable.
2020NeurIPSno evidence yet
0124Scaling Laws for Neural Language Models
Loss as a power law of compute, data, parameters; the industry's planning document.
2020arXivno evidence yet
0125Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
RAG; grounding generation in retrieved evidence.
2020NeurIPSno evidence yet
0126Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
MuZero; planning without knowing the rules.
2020Natureno evidence yet
0127Zoom In: An Introduction to Circuits
Founded mechanistic interpretability: studying networks like organisms.
2020Distillno evidence yet
0128Closing the AI Accountability Gap: Defining an End-to-End Framework for Internal Algorithmic Auditing
The reference framework for internal AI audit practice.
2020FAT*no evidence yet
0129An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
ViT; transformers displace convolutions in vision.
2021ICLRno evidence yet
0130Learning Transferable Visual Models From Natural Language Supervision
CLIP; vision-language alignment underpinning multimodal AI.
2021ICMLno evidence yet
0131Evaluating Large Language Models Trained on Code
Codex; LLMs write code, the capability that transformed software work.
2021arXivno evidence yet
0132Unsolved Problems in ML Safety
The mainstream-ML framing of robustness, monitoring, alignment, systemic safety.
2021arXivno evidence yet
0133Datasheets for Datasets
Provenance documentation for training data.
2021Communications of the ACMno evidence yet
0134On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
The defining critique of the LLM paradigm, and the paper behind Google's Gebru affair.
2021FAccTno evidence yet
0135Extracting Training Data from Large Language Models
LLMs memorize and can regurgitate training data.
2021USENIX Securityno evidence yet
0136Highly Accurate Protein Structure Prediction with AlphaFold
Solved a fifty-year grand challenge; the Nobel-recognized proof of AI for science.
2021Natureno evidence yet
0137The Hardware Lottery
Research directions win because hardware favors them; a structural critique of progress.
2021Communications of the ACMno evidence yet
0138High-Resolution Image Synthesis with Latent Diffusion Models
Stable Diffusion; open-weights image generation at consumer scale.
2022CVPRno evidence yet
0139Training Compute-Optimal Large Language Models
Chinchilla; rebalanced the field toward data-optimal training.
2022NeurIPSno evidence yet
0140LoRA: Low-Rank Adaptation of Large Language Models
Made finetuning large models affordable; standard adaptation method.
2022ICLRno evidence yet
0141Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity
Mixture-of-experts at scale; the sparse path to frontier capability.
2022Journal of Machine Learning Researchno evidence yet
0142Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Stepwise prompting unlocks latent reasoning; opened the reasoning-model agenda.
2022NeurIPSno evidence yet
0143Emergent Abilities of Large Language Models
Named (and contested) the phenomenon of capability jumps with scale.
2022TMLRno evidence yet
0144Training Language Models to Follow Instructions with Human Feedback
InstructGPT; RLHF at scale, the technique behind aligned chat models.
2022NeurIPSno evidence yet
0145Constitutional AI: Harmlessness from AI Feedback
Alignment via explicit principles and AI feedback; reduced dependence on human labeling.
2022arXivno evidence yet
0146Toy Models of Superposition
Why features share neurons; the core obstacle to reading networks.
2022Anthropic / Transformer Circuitsno evidence yet
0147Discovering Faster Matrix Multiplication Algorithms with Reinforcement Learning
AlphaTensor; AI finds new mathematics.
2022Natureno evidence yet
0148Segment Anything
Promptable foundation model for segmentation.
2023ICCVno evidence yet
0149ReAct: Synergizing Reasoning and Acting in Language Models
Reason-act loops; the template for LLM agents.
2023ICLRno evidence yet
0150Toolformer: Language Models Can Teach Themselves to Use Tools
Self-supervised tool use; agents calling APIs.
2023NeurIPSno evidence yet
0151LLaMA: Open and Efficient Foundation Language Models
The open-weights line that created today's open-model ecosystem.
2023arXivno evidence yet
0152GPT-4 Technical Report
The frontier capability report; also the moment training details went dark.
2023arXivno evidence yet
0153Sparks of Artificial General Intelligence: Early Experiments with GPT-4
The most-debated capability claim of the era; framed the AGI-proximity argument.
2023arXivno evidence yet
0154Mamba: Linear-Time Sequence Modeling with Selective State Spaces
The leading post-Transformer architecture candidate.
2023arXivno evidence yet
0155Universal and Transferable Adversarial Attacks on Aligned Language Models
Automated jailbreaks; alignment as an attack surface.
2023arXivno evidence yet
0156Towards Monosemanticity: Decomposing Language Models with Dictionary Learning
Sparse autoencoders extract human-legible features from LLMs.
2023Anthropic / Transformer Circuitsno evidence yet
0157Frontier AI Regulation: Managing Emerging Risks to Public Safety
The reference proposal for frontier-model regulatory architecture.
2023arXivno evidence yet
0158GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models
First systematic occupational-exposure estimates for LLMs.
2023arXiv (later Science)no evidence yet
0159Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training
Empirical evidence that deceptive behaviour can survive standard safety training.
2024arXivno evidence yet
0160Managing Extreme AI Risks Amid Rapid Progress
Consensus statement by senior researchers on frontier-risk preparedness.
2024Scienceno evidence yet
0161Accurate Structure Prediction of Biomolecular Interactions with AlphaFold 3
Extended structure prediction to complexes and drug-relevant interactions.
2024Natureno evidence yet
0162DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
Open reasoning model from China that reset assumptions about cost and access.
2025arXivno evidence yet