Home / Publications
Publications
62 selected papers. The complete list is on Google Scholar (10,043 citations, h-index 44, October 2026).
2026
Decocted Experience Improves Test-Time Inference in LLM Agents
Scales an agent’s context instead of its compute. Experience distilled into lessons, consolidated and retrieved beats raw experience on math, web and software tasks.
Building a Foundational Guardrail for General Agentic Systems via Synthetic Data
A guardrail that runs before an agent acts. AuraGen synthesizes risky plans, Safiron flags and explains them, Pre-Exec Bench measures both.
Answering the Wrong Question: Reasoning Trace Inversion for Abstention in LLMs
Reconstructs the question from a model’s reasoning trace; if it drifts from the question asked, abstain. Best in 33 of 36 settings.
Multi-component Causal Tracing in Large Language Models
Finds the set of attention heads and MLP neurons that jointly drive a metric, by relaxing the combinatorial search to a continuous one.
TrustGen: A Platform of Dynamic Benchmarking on the Trustworthiness of Generative Foundation Models
BLINDSPOT: A Benchmark for Safety and Refusal Calibration in Long-Horizon Tool-Using Agents
Judges whole agent trajectories, not turns: 2,500+ runs averaging 14.7 turns, each labeled safe, correctly refused, unsafe, over-refused or unclear.
LCGuard: Latent Communication Guard for Safe KV Sharing in Multi-Agent Systems
Agents that share KV caches leak private context. A learned transform of the cache keeps the task signal and blocks reconstruction.
2025
Granite Guardian: Comprehensive LLM Safeguarding
One model for harm, jailbreaks and RAG hallucination, trained on human labels plus synthetic data. AUC 0.871 and 0.854 on the two benchmark families.
Multi-Level Explanations for Generative Language Models
LIME- and SHAP-style attribution for long-context generation, at a cost that scales. More faithful than LLM self-explanations in human and automated tests.
When in Doubt, Cascade: Towards Building Efficient and Capable Guardrails
Traced a bias detector’s errors to the use–mention distinction, then generated 300K contrastive examples to fix it at a fraction of the compute.
Contextual Value Alignment
Adversarial Prompt Evaluation: Systematic Benchmarking of Guardrails Against Prompt Input Attacks on LLMs
2024
Thermometer: Towards Universal Calibration for Large Language Models
An auxiliary model, trained across tasks, that calibrates an LLM on tasks it has not seen. Accuracy unchanged, inference cost negligible.
Graph-based Uncertainty Metrics for Long-form Language Model Generations
Treats samples and their claims as a bipartite graph; claim uncertainty is graph centrality. Self-consistency is the degree-centrality special case.
Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage?
Shows evidential deep learning’s epistemic uncertainty does not behave as claimed, and that the methods are better read as energy-based OOD detectors.
WikiContradict: A Benchmark for Evaluating LLMs on Real-World Knowledge Conflicts from Wikipedia
Interventional Causal Discovery in a Mixture of DAGs
Language Models in Dialogue: Conversational Maxims for Human-AI Interactions
Value Alignment from Unstructured Text
Causal Bandits with General Causal Models and Interventions
ComVas: Contextual Moral Values Alignment System
Large Language Model Confidence Estimation via Black-Box Access
The RealHumanEval: Evaluating Large Language Models’ Abilities to Support Programmers
Alignment Studio: Aligning Large Language Models to Particular Contextual Regulations
Fourier Neural Operators for Arbitrary Resolution Climate Data Downscaling
Separability Analysis for Causal Discovery in Mixture of DAGs
2023
Effective Human-AI Teams via Learned Natural Language Rules and Onboarding
Finds where people over- or under-trust a model, describes each region in plain language, and teaches it. About 5% more accurate human–AI teams.
Who Should Predict? Exact Algorithms for Learning to Defer to Humans
Learning a linear classifier–rejector pair is NP-hard even when a perfect one exists. An exact MILP for small cases, a consistent surrogate for the rest.
Efficient Equivariant Transfer Learning from Pretrained Models
Makes a pretrained model equivariant by averaging over group-transformed features, with the weights learned rather than fixed.
Post-hoc Uncertainty Learning Using a Dirichlet Meta-Model
Equi-Tuning: Group Equivariant Fine-Tuning of Pretrained Models
Reliable Gradient-free and Likelihood-free Prompt Tuning
Add-Remove-or-Relabel: Practitioner-Friendly Bias Mitigation via Influential Fairness
Hard-Constrained Deep Learning for Climate Downscaling
Causal Bandits for Linear Structural Equation Models
The Incentive Gap in Data Work in the Era of Large Models
2022
Fair Infinitesimal Jackknife: Mitigating the Influence of Biased Training Data Points Without Refitting
Uses infinitesimal-jackknife influence to find the training points that hurt a fairness metric, and removes their effect without refitting.
Selective Regression Under Fairness Criteria
Intervention Target Estimation in the Presence of Latent Variables
Causal Feature Selection for Algorithmic Fairness
AI Explainability 360: Impact and Design
2021
Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty
Treats communicated uncertainty as a form of transparency: how to measure it, show it, and act on it with stakeholders.
Fair Selective Classification via Sufficiency
Scalable Intervention Target Estimation in Linear Models
AdaFuse: Adaptive Temporal Fusion Network for Efficient Action Recognition
Leveraging Latent Features for Local Explanations
Conditionally Independent Data Generation
2020
not-so-BigGAN: Generating High-Fidelity Images on Small Compute with Wavelet-based Super-Resolution
Samples in the wavelet domain and super-resolves to pixels. FID 10.59 on ImageNet 512 at half BigGAN’s compute.
Optimizing Mode Connectivity via Neuron Alignment
AR-Net: Adaptive Frame Resolution for Efficient Action Recognition
TAFSSL: Task-Adaptive Feature Sub-Space Learning for Few-Shot Classification
Fairness of Classifiers Across Skin Tones in Dermatology
Building Calibrated Deep Models via Uncertainty Matching with Auxiliary Interval Predictors
2019
Learning New Tricks From Old Dogs: Multi-Source Transfer Learning From Pre-Trained Networks
AI Fairness 360: An Extensible Toolkit for Detecting and Mitigating Algorithmic Bias
Fairness GAN: Generating Datasets with Fairness Properties Using a Generative Adversarial Network
An auxiliary-classifier GAN that generates data close to the original but satisfying demographic parity or equal opportunity.
Understanding Behavior of Clinical Models under Domain Shifts
2018
Variational Inference of Disentangled Latent Concepts from Unlabeled Observations
A regularizer on the aggregate posterior that yields disentangled latents without labels, plus the SAP score to measure it.
Co-regularized Alignment for Unsupervised Domain Adaptation
Aligns source and target in several diverse feature spaces at once and makes their target predictions agree.
2017
Semi-supervised Learning with GANs: Manifold Invariance with Improved Inference
Uses the GAN generator to estimate the data manifold’s tangent space, then makes the classifier invariant along it.