FixMatch - Simplifying Semi-Supervised Learning with Consistency and Confidence
Explore FixMatch, a powerful semi-supervised learning approach combining consistency regularization and pseudo-labeling for improved model performance with limited labeled data.
Dynamical Distance Learning for Semi-Supervised and Unsupervised Skill Discovery
Explore dynamical distance learning for reinforcement agents, enhancing policy learning through state distance estimation. Discover applications in semi-supervised and unsupervised skill acquisition.
CURL - Contrastive Unsupervised Representations for Reinforcement Learning
Explore how contrastive learning enhances reinforcement learning, improving performance in complex tasks using high-level features extracted from raw pixels.
POET - Endlessly Generating Increasingly Complex and Diverse Learning Environments and Solutions
Explore POET, an innovative algorithm that generates and solves increasingly complex challenges, fostering open-ended learning and diverse problem-solving capabilities through evolutionary methods and transfer learning.
Dream to Control - Learning Behaviors by Latent Imagination
Explore DeepMind's Dreamer, an innovative RL agent that learns complex behaviors through latent imagination, surpassing existing approaches in visual control tasks.
Agent57 - Outperforming the Atari Human Benchmark
Explore DeepMind's Agent57, the first AI to surpass human performance in all 57 Atari games. Learn about its innovative approach to reinforcement learning and meta-learning for exploration-exploitation balance.
Deep Learning for Symbolic Mathematics
Explore neural networks solving complex mathematical problems like symbolic integration and differential equations, outperforming traditional Computer Algebra Systems.
Turing-NLG, DeepSpeed and the ZeRO Optimizer
Explore Microsoft's 17-billion parameter language model, ZeRO optimizer, and DeepSpeed, enabling efficient model and data parallelism for state-of-the-art natural language processing breakthroughs.
Reformer - The Efficient Transformer
Discover how Reformer revolutionizes Transformer models, reducing memory usage and enabling processing of longer sequences through innovative techniques like Locality Sensitive Hashing and Reversible Networks.
Reinforcement Learning Upside Down - Don't Predict Rewards, Just Map Them to Actions
Innovative approach to reinforcement learning that maps desired rewards to actions, transforming it into supervised learning. Shows promising performance compared to traditional RL algorithms.
A Neurally Plausible Model Learns Successor Representations in Partially Observable Environments
Explores a neurally plausible model using distributional successor features for efficient reinforcement learning in partially observable, noisy environments, bridging model-based and model-free approaches.
AlphaStar - Grandmaster Level in StarCraft II Using Multi-Agent Reinforcement Learning
Explore DeepMind's AI agent achieving Grandmaster-level skill in StarCraft II through innovative multi-agent reinforcement learning and League Training techniques.
IMPALA - Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
Explore IMPALA, a distributed reinforcement learning architecture for efficient multi-task learning. Learn about V-trace, off-policy correction, and scalable performance across thousands of machines.
Dynamic Routing Between Capsules
Explore Capsule Networks, a novel neural network architecture using capsules and dynamic routing for improved object recognition and handling of spatial relationships.
Manifold Mixup: Better Representations by Interpolating Hidden States
Explore Manifold Mixup, a regularization technique that improves neural network performance by interpolating hidden representations, leading to smoother decision boundaries and more robust predictions.
Conversation About Population-Based Methods
Explore population-based methods and open-ended learning in AI through an insightful interview, delving into cutting-edge research and practical applications in machine learning.
Adversarial Examples Are Not Bugs, They Are Features
Explore the concept of adversarial examples in machine learning, their origins, and implications for AI robustness and human-AI alignment.
Reinforcement Learning, Fast and Slow
Explores recent advancements in deep reinforcement learning, discussing techniques for faster learning and their potential implications for cognitive science and neuroscience.
The Odds are Odd: A Statistical Test for Detecting Adversarial Examples
Explore statistical methods for detecting adversarial examples in machine learning models, focusing on test statistics and their effectiveness in identifying manipulated inputs.
GPT-2: Language Models are Unsupervised Multitask Learners
Explore OpenAI's GPT-2 model, its capabilities in natural language processing tasks, and the implications of its performance on various datasets without explicit supervision.
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Explore how Batch Normalization accelerates deep network training by reducing internal covariate shift, enabling higher learning rates and improved model performance.