Research Hall of Fame
Important papers from the history of AI, computing, and cognitive science, explained in plain English.
Browse topics
Agents
Building Autonomous Lifelong Learning Agents
Voyager is an LLM-powered embodied agent that uses iterative code generation and a skill library to perform continuous autonomous learning in open-ended environments.
2023Building Autonomous Agents with Social Believability
The researchers developed an architecture that enables autonomous agents to simulate human behavior, including daily activities, social interactions, and long-term memory, using a large language model.
2023Teaching Language Models To Use Tools
Toolformer enables language models to autonomously determine when and how to call external APIs to overcome limitations in arithmetic and factual accuracy.
2022Improving AI Decision Making Through Reasoning
The ReAct framework enhances language model decision making by interleaving logical reasoning traces with external tool actions.
Artificial Intelligence
Predicting Protein Structures Using Deep Learning
AlphaFold utilizes a novel deep learning architecture to predict three-dimensional protein structures from amino acid sequences with accuracy competitive with experimental methods.
2020Few Shot Learning with Large Language Models
The researchers introduced GPT-3, an autoregressive language model that performs tasks through text interaction without needing task-specific fine-tuning.
2018Improving Language Understanding with Bidirectional Transformers
The paper introduces BERT, a pre-trained language model that uses bidirectional context to improve performance on various sentence-level and token-level tasks.
2017Replacing Recurrent Networks with Attention
The Transformer architecture replaces sequential recurrent and convolutional layers with an attention-based mechanism that enables massive parallelization for sequence processing.
2014An Efficient Optimization Method for Training
The paper introduces Adam, an efficient gradient based optimization method designed to handle large scale stochastic objective functions with noisy or sparse gradients.
2014Generative Adversarial Networks for Data Modeling
The paper introduces a framework to train generative and discriminative models simultaneously as a minimax two-player game to better estimate data distributions.
2013Efficiently Computing Vector Representations of Words
The paper introduces two new model architectures that compute continuous word vector representations with improved computational efficiency and accuracy compared to existing neural network techniques.
1997Introducing Long Short-Term Memory Networks
The paper introduces a gradient-based method called Long Short-Term Memory that enables neural networks to store and recall information over long sequences.
Benchmarks & Evals
Evaluating Language Models on Real Software Issues
The researchers developed SWE-bench to measure how effectively language models can resolve complex software engineering problems within real-world GitHub repositories.
2021Evaluating Large Language Models Trained on Code
The paper evaluates how well large language models synthesize computer programs from natural language docstrings.
2020Measuring Language Models on Broad Knowledge
The researchers developed a standardized benchmark of 57 academic and professional tasks to evaluate the multitask accuracy of large language models.
2018Introducing the GLUE Benchmark for NLU
The authors created the GLUE benchmark as a standardized tool to evaluate and improve the performance of natural language understanding systems across diverse tasks.
2009Building a Large Scale Image Database
The authors created a massive image database by organizing internet images according to the semantic structure of WordNet.
Computer Vision
Introducing the Segment Anything Model
The researchers developed a promptable model and a massive dataset to enable zero-shot segmentation for computer vision tasks.
2021Efficient Image Synthesis Using Latent Diffusion Models
The researchers developed Latent Diffusion Models (LDMs) that move image generation from pixel space to a compressed latent space to reduce training and inference costs.
2020Using Transformers for Computer Vision Tasks
This research applies the Transformer architecture directly to image recognition by treating image patches as a sequence of data, proving it can rival existing convolutional networks.
2020High Quality Image Synthesis Using Diffusion
The researchers introduce a high quality image synthesis method using diffusion probabilistic models inspired by nonequilibrium thermodynamics.
2015Training Deeper Neural Networks Using Residuals
The researchers introduce a residual learning framework to simplify the training of significantly deeper neural networks.
2015Real-Time Object Detection with YOLO
The paper introduces a unified neural network architecture that treats object detection as a single regression problem to achieve high-speed, real-time performance.
2015Efficient Biomedical Image Segmentation Architecture
The researchers introduced U-Net, a symmetric convolutional network designed for high-accuracy pixel-level image segmentation using limited training data.
2012Deep Learning for Large Scale Image Classification
The authors developed a deep convolutional neural network that significantly improved classification performance on the ImageNet dataset by leveraging GPU acceleration and new regularization techniques.
Efficiency & Inference
Accelerating Transformer Inference Through Speculative Decoding
The authors introduce a method to speed up large language model inference by using a smaller model to generate draft tokens that a larger model then verifies in parallel.
2022Efficient 8-bit Matrix Multiplication for Transformers
The researchers introduced LLM.int8(), a method that reduces the GPU memory required for large language model inference by 50 percent while maintaining full precision performance.
2022Making Transformer Attention Faster and Efficient
FlashAttention improves the speed and memory efficiency of Transformers by using tiling to optimize read and write operations between GPU memory levels.
2018Finding Trainable Sparse Neural Networks
The researchers discovered that dense neural networks contain smaller, sparse subnetworks that can be trained effectively from their original random initializations.
2017Scaling Neural Networks with Sparse Experts
The researchers implemented a Sparsely-Gated Mixture of Experts layer to increase neural network capacity to 137 billion parameters without a proportional increase in computational cost.
2015Compressing Neural Networks Using Knowledge Distillation
The authors present a method for compressing the knowledge held by a large ensemble of neural networks into a single, smaller, and more efficient model.
Multimodal
Building Visual Assistants with Instruction Tuning
The paper introduces LLaVA, a general-purpose visual assistant built using language-only GPT models and visual encoders to follow multimodal vision-and-language instructions.
2022Robust Speech Recognition via Weak Supervision
The paper introduces a robust speech processing approach using large-scale weak supervision to generalize across benchmarks without domain-specific fine-tuning.
2022Flamingo: Few-Shot Visual Language Models
Flamingo is a visual language model designed to perform diverse multimodal tasks by learning from a few examples without requiring task-specific fine-tuning.
2021Learning Visual Models Using Natural Language
The researchers developed CLIP, a model trained on 400 million image-text pairs to identify visual concepts using natural language instead of fixed category labels.
2021Zero-Shot Generation of Images from Text
The paper presents a text-to-image generation method that removes the need for complex, auxiliary modeling assumptions by using an autoregressive transformer approach.
Reasoning
Teaching Language Models to Reason via Reinforcement Learning
The paper introduces DeepSeek-R1, a model trained using pure reinforcement learning to develop reasoning capabilities without relying on human-annotated demonstrations.
2023Improving Model Reasoning With Stepwise Feedback
The researchers demonstrate that providing feedback on each intermediate reasoning step instead of just the final result significantly improves the success rate of large language models on complex tasks.
2023Structuring LLM Thinking Through Tree Search
The paper introduces the Tree of Thoughts (ToT) method, which allows language models to perform strategic planning and lookahead by exploring multiple reasoning paths instead of a single linear chain.
2022Bootstrapping Language Model Reasoning Capabilities
STaR improves model reasoning by iteratively generating rationales, correcting incorrect answers, and fine-tuning on successful logic paths.
2022Improving LLM Reasoning Through Self-Consistency
The paper introduces a decoding strategy that samples multiple reasoning paths and selects the most consistent answer to improve performance on complex tasks.
2022Improving Large Language Model Reasoning Capability
The researchers show that prompting large language models with a sequence of intermediate reasoning steps significantly improves their performance on complex tasks.
Reinforcement Learning
Efficient Policy Optimization for RL Agents
The paper introduces Proximal Policy Optimization, a simplified algorithm that allows for more sample-efficient training compared to traditional policy gradient methods.
2016Mastering the Game of Go with AI
The paper introduces AlphaGo, a system that uses deep neural networks and tree search to solve the complex game of Go by learning from both human expert games and self-play.
2013Learning Atari Game Policies from Pixels
The researchers developed a system that learns to play Atari games by processing raw screen pixels directly through a neural network.
Robotics
Teaching Robots Using Internet Vision Knowledge
The RT-2 model improves robot control by fine-tuning vision-language models on robotic trajectory data alongside large-scale web-based datasets.
2022A Robotics Transformer for Real World Control
RT-1 is a robotic control model that leverages transformer architecture to process visual and language data for real-time task execution.
2015End to End Robot Control Learning
The paper investigates if training perception and control systems together as a single unit performs better than building them as separate components.
Safety & Alignment
Training Harmless AI Using AI Feedback
The paper presents a method for training harmless AI assistants without relying on human labels to identify harmful outputs.
2022Training Language Models Using Human Feedback
The researchers aligned language models with human intent by using supervised learning combined with reinforcement learning from human feedback.
2017Teaching AI With Human Feedback
The paper introduces a method to train reinforcement learning systems using human preferences instead of a traditional reward function.
2016Practical Approaches to Preventing AI Accidents
The paper identifies and categorizes five specific technical problem areas that cause unintended and harmful behaviors in machine learning systems.
Training & Fine-Tuning
Direct Preference Optimization for Language Models
The paper introduces Direct Preference Optimization as a simpler method to align language models with human preferences without using reinforcement learning.
2022Optimizing Data for Large Language Models
The researchers demonstrate that transformer models are often undertrained and achieve better performance by scaling training data in proportion to model size.
2021Efficient Fine-Tuning for Large Language Models
LoRA is a technique that enables fine-tuning of large models by injecting small trainable matrices into Transformer layers instead of updating the entire model weight set.
2020Scaling Laws for Neural Language Models
Researchers established empirical scaling laws demonstrating that language model performance improves predictably with increases in model size, dataset size, and compute budget.
2018Universal Fine-Tuning for Text Classification
The researchers developed ULMFiT to bring the efficiency of transfer learning to text classification tasks.