Virginia Smith
Associate Professor
Studies large-scale and federated machine learning, optimization, robustness and privacy.
Virginia Smith's lab at CMU works on trustworthy & safe ML and learning theory & optimization. 41 papers since 2023, 27 at top venues, mostly NeurIPS and ICLR. 11 alumni. Applications close .
At a glance
- papers, 2023–2026
- 11/ year
- at T1 venues
- 66%
- PhD students
- 0
- MS students
- 0
- alumni listed
- 11
Papers by yearfrom the lab homepage, 2023–2026
- 12 papers2026 · T1 6 · collecting
- 11 papers2025 · T1 6
- 9 papers2024 · T1 6
- 9 papers2023 · T1 9
T1 venuesOtherVenues: NeurIPS, ICLR, TMLR, ICML, EMNLP
Research focuspaper titles that use each topic's words
- Learning theory & optimization11 papers
- federated6 papers
- Trustworthy & safe ML1 paper
- studies large-scale0 papers
- federated machine0 papers
Paper list
From the lab homepage.
202612 papers
- EMNLP (T1)DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment
- ICLR (T1)e3: Learning to Explore Enables Extrapolation of Test-Time Compute for LLMs
- SaTMLExact Unlearning of Finetuning Data via Model Merging at Scale
- CAISFederated LoRA with Sparse Communication
- ICML (T1)IsoCompute Playbook: Optimally Scaling Sampling Compute for LLM RL
- ACL (T1)Mechanistic Interpretability Should Prioritize Feature Consistency in SAEs
- ICML (T1)Membership Inference Attacks for Unseen Classes
- COLMPando: Do Interpretability Methods Work When Models Won't Explain Themselves?
- TMLR (T1)POPE: Learning to Reason on Hard Problems via Privileged On-Policy Exploration
- International Conference on Machine Learning (ICML) Position Paper TrackPreventing AI-Generated CSAM Necessitates New Approaches to AI Safety
- EACLRefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models
- SaTMLResearch in Collaborative Learning Does Not Serve Cross-Silo Federated Learning in Practice
202511 papers
- TMLR (T1)Agreement-Based Cascading for Efficient Inference
- —BLUR: A Benchmark for LLM Unlearning Robust to Forget-Retain Overlap
- SOSP (T1)COpter: Efficient Large-Scale Resource-Allocation via Continual Optimization
- NAACL (T1)CoRAG: Collaborative Retrieval-Augmented Generation
- NAACL (T1)Decoding Dark Matter: Specialized Sparse Autoencoders for Interpreting Rare Concepts in Foundation Models
- SaTMLLLM Unlearning Benchmarks are Weak Measures of Progress
- ICLR (T1)Many-Objective Multi-Solution Transport
- Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks TrackPARALLELPROMPT: Extracting Parallelism from Large Language Model Queries
- COLMSAEs Can Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMs
- Empirical Methods in Natural Language Processing (EMNLP) Industry TrackSemantic Agreement Enables Efficient Open-Ended LLM Cascades
- ICLR (T1)Unlearning or Obfuscating? Jogging the Memory of Unlearned LLMs via Benign Relearning
20249 papers
- —Everybody Prune Now: Structured Pruning of LLMs with only Forward Passes
- SaTMLFair Federated Learning via Bounded Group Loss
- EMNLP (T1)GRASS: Compute Efficient Low-Memory LLM Training with Structured Sparse Gradients
- ICLR WorkshopGuardrail Baselines for Unlearning in LLMs
- TMLR (T1)Maximizing Global Model Appeal in Federated Learning
- NeurIPS (T1)No Free Lunch in LLM Watermarking: Trade-offs in Watermarking Design Choices
- NeurIPS (T1)On the Benefits of Public Representations for Private Transfer Learning under Distribution Shift
- ICML (T1)Prompting is a Double-Edged Sword: Improving Worst-Group Robustness of Foundation Models
- NeurIPS (T1)RL on Incorrect Synthetic Data Scales the Efficiency of LLM Math Reasoning by Eight-Fold
20239 papers
- ICLR (T1)Bitrate-Constrained DRO: Beyond Worst Case Robustness To Unknown Group Shifts
- NeurIPS (T1)Complementary Benefits of Contrastive Learning and Self-Training Under Distribution Shift
- ICLR (T1)Differentially Private Adaptive Optimization with Delayed Preconditioners
- MLSys (T1)On Noisy Evaluation in Federated Hyperparameter Tuning
- JMLR (T1)On Tilted Losses in Machine Learning: Theory and Applications
- TMLR (T1)Private Multi-Task Learning: Formulation and Applications to Federated Learning
- NeurIPS (T1)Progressive Knowledge Distillation: Building Ensembles for Efficient Inference
- MLSys (T1)Validating Large Language Models with ReLM
- NeurIPS (T1)Variance-Reduced Gradient Estimation via Noise-Reuse in Online Evolution Strategies
Where alumni went11 listed, 0% to industry
Industry 0Academia 5
Also to
- Stanford University
- Carnegie Mellon University
- UIUC
Apply to CMUFall 2027
CMU SCS Computer Science PhD
- Round
- Single round
- Opens
- GRE
- See program page
- English
- TOEFL / IELTS / Duolingo (F-1/J-1 non-native speakers)
- Fee
- USD 100 (USD 80 by early deadline)
Confirmed on the program page.
Similar labsnearest by research summary
- CNSJTU
Wenjun Zhang
- Image & video generation
- USNYU
Shengjie Wang
- Learning theory & optimization
- USUC Berkeley
Song Mei
- Learning theory & optimization
- CNPeking
Tongyang Li
- Learning theory & optimization
- USUCSD
Daniel M. Kane
- Learning theory & optimization
- USUC Berkeley
Somayeh Sojoudi
- Trustworthy & safe ML
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
CMUcloses