Safety

A model that behaves safely in evaluation can stop behaving safely once it is fine-tuned, narrowed to one job, or pushed by an adversary. We study whether safety survives those conditions: whether a task-specific agent refuses what falls outside its remit, whether alignment persists after adaptation, and how it can be restored or steered at test time without giving up capability.

Operational safety

OffTopicEval

An evaluation suite measuring whether LLM agents accept valid in-domain requests and refuse out-of-domain ones.

Alignment

Safety Arithmetic

A test-time framework for steering language models toward safer behavior through parameters and activations.

Re-alignment

RESTA

Restoring safety in fine-tuned language models through task arithmetic while retaining downstream capability.

Red-teaming

Chain of Utterances

RED-EVAL and Chain-of-Utterances prompting for probing harmful behavior and studying safety alignment.

Bias analysis

Gender Bias in BERT

A highly cited analysis of gender bias encoded in contextualized language representations.

Trustworthiness

A fluent answer and a trustworthy one are not the same thing. We work on systems that ground what they say in retrieved evidence, attribute it honestly, decline when the evidence will not support an answer, and — where several agents cooperate — calibrate how much they should rely on one another.

Retrieval-augmented generation

Trust-Score and Trust-Align

Trust-Score evaluates RAG trustworthiness; Trust-Align improves grounded attribution, refusal, and citation quality.

Grounded generation

Chain-of-Knowledge

A highly cited framework for grounding generation through dynamic knowledge adaptation over heterogeneous sources.

Multi-agent trust

Epistemic Context Learning

Trust formation and calibrated reliance in LLM-based multi-agent systems.

Multimodality

Meaning arrives through words, voice, face, and scene at once, and a model that reads only one channel misses most of it. We study how to fuse those channels — separating what they share from what belongs to each — and how to generate in them, from emotion in multi-party conversation to text-to-audio synthesis.

Conversational AI

MELD and DialogueRNN

Highly cited resources and models for multimodal, multi-party emotion recognition in conversations.

Representation and fusion

TFN, MISA, and Multimodal-Infomax

Foundational work on modality interaction, invariant and specific representations, and robust multimodal sentiment analysis.

Audio generation

Tango and TangoFlux

Text-to-audio generation spanning diffusion-based synthesis and fast preference-optimized flow matching.

AI for Science

We ask whether language models can do more than summarize the literature — whether they can propose hypotheses a scientist would think worth testing. Measuring that claim honestly is half the problem, so we build benchmarks that check whether a model rediscovers findings deliberately held out of its training data.

Scientific discovery

MOOSE-Chem

A benchmark and framework testing whether LLMs can rediscover valid chemistry hypotheses from background literature.

Hypothesis discovery

Open-Domain Scientific Hypotheses

Methods for automated scientific hypothesis discovery across broad literature collections.

Efficiency

Capability that is too expensive to train, serve, or remember does not reach anyone. We work on getting more from a fixed budget: compact online memory instead of ever-longer context, learned data selection instead of simply more data, merging instead of retraining, and exact linear-time attention for long inputs.

Online memory

δ-mem

A lightweight online memory that compresses history into a compact state and uses it to modulate a frozen model.

Data selection

Data Agent

End-to-end dynamic data selection that learns sample-wise policies to accelerate training while preserving accuracy.

Model merging

DELLA-Merging

Magnitude-based sampling to reduce interference when merging task-specialized language models.

Linear attention

EFLA

Error-Free Linear Attention derives an exact continuous-time solution for robust long-context computation.

Efficient adaptation

PromptDistill, LLM-Adapters, UDApter

Efficient inference, parameter-efficient fine-tuning, and adapter-based transfer for language and speech models.

Embodied AI

Language and vision only matter to a robot if they end in the right action. We build compact vision-language-action models that ground perception in behavior, and evaluations that test the whole path — from what the agent sees, through what it decides, to what it actually does.

Vision-language-action

NORA and NORA 1.5

Small generalist vision-language-action models for efficient action grounding and dependable embodied behavior.

Interactive evaluation

Emma-X and Perception-to-Action

Embodied foundation models and benchmarks evaluating how agents move from visual reasoning to action.