Researchers

Learning useful internal representations from data.

Learning distributed representations that can support generalization.

Learning useful visual features and predictive representations.

Using large neural networks to learn sequence transformations.

Training a generator through competition with a discriminator.

Connecting visual recognition, language, and embodied intelligence.

Learning features from unlabeled data at scale.

Combining learning and search to solve structured problems.

Learning decisions through interaction and search.

Learning predictions and actions from reward and experience.

Using probabilistic structure to learn from complex data.

Learning latent-variable models with efficient gradient estimators.

Using probabilistic latent spaces to model observations.

Making optimization and attention more efficient.

Connecting neural language generation with visual understanding.

Building the systems that make large-scale learning practical.

Learning sequence representations and scaling neural networks.

Treating structured predictions as sequences.

Studying scaling and the reliability of learned systems.

Measuring how loss changes with model size, data, and compute.

Studying predictable relationships between resources and learning.

Learning transferable representations through large-scale pretraining.

Improving policies while controlling the size of each update.

Studying helpfulness, honesty, and harmlessness in AI assistants.

Investigating the internal features and computations of networks.

Combining attention and sparse conditional computation.

Using attention to process sequences in parallel.

Connecting linguistic structure with learned language representations.

Making language-model evaluation broader and more transparent.

Learning adaptable skills from limited experience.

Learning policies for difficult perception and control problems.

Learning robot behavior from data with deep reinforcement learning.

Making deep visual models easier to optimize and transfer.

Studying deep architectures for visual and language modeling.

Generating data by learning to reverse a noising process.

Learning score functions to guide generative sampling.

Connecting probabilistic modeling, optimization, and generation.

Reducing memory traffic and computation in sequence models.

Making learning systems more efficient and usable.

Modeling long sequences with structured state spaces.

Compressing and accelerating neural networks for deployment.

Reducing the memory cost of large-model training and inference.

Learning language systems that can be adapted efficiently.

Combining retrieval with language understanding.

Eliciting intermediate reasoning from language models.

Studying compute-efficient training and sparse language models.

Studying reproducibility and evaluation in machine learning.

Studying commonsense knowledge and model behavior.

Learning operators and representations for scientific problems.

Controlling image synthesis through style-based generator architectures.