Tsui-Wei Weng
Assistant Professor
Studies robustness, interpretability and safety of deep neural networks, including certified robustness and automated neuron explanation.
Tsui-Wei Weng's lab at UCSD works on trustworthy & safe ML, interpretability, evaluation & safety of LMs and data management & big data. 43 papers since 2023, 32 at top venues, mostly ICML and NeurIPS. 4 PhD students and 2 master's students. 4 alumni, most often to Amazon and Apple. Applications close .
At a glance
- papers, 2023–2026
- 11/ year
- at T1 venues
- 74%
- PhD students
- 4
- MS students
- 2
- alumni listed
- 4
Papers by yearfrom the lab homepage, 2023–2026
- 11 papers2026 · T1 8 · collecting
- 10 papers2025 · T1 8
- 13 papers2024 · T1 11
- 9 papers2023 · T1 5
T1 venuesOtherVenues: ICML, NeurIPS, ICLR, TMLR, CVPR
Research focuspaper titles that use each topic's words
- Interpretability, evaluation & safety of LMs5 papers
- Data management & big data3 papers
- Trustworthy & safe ML2 papers
- robustness2 papers
- explanation0 papers
Paper list
From the lab homepage.
202611 papers
- CVPR (T1)Beyond Top Activations: Efficient and Reliable Crowdsourced Evaluation of Automated Interpretability
- TMLR (T1)CI-CBM: Class-Incremental Concept Bottleneck Model for Interpretable Continual Learning
- COLMData-Efficient Adaptation of LLMs via Attention Head Reweighting
- TMLR (T1)Graph Concept Bottleneck Models
- CVPR (T1)Interpretable and Steerable Concept Bottleneck Sparse Autoencoders
- NeurIPS (T1)LLM Agents Already Know When to Call Tools - Even Without Reasoning
- COLMReFIne: A Framework for Trustworthy Large Reasoning Models with Reliability, Faithfulness, and Interpretability
- COLM (accepted), NeurIPS Mechanistic Interpretability WorkshopReflCtrl: Controlling LLM Reflection Efficiently via Representation Engineering
- ICML (T1)Resting Neurons, Active Insights: Robustifying Activation Sparsity in LLMs via Spontaneity
- NeurIPS (T1)Steer2Edit: From Activation Steering to Component-Level Editing
- EMNLP (T1)The Cold-Start Safety Gap in LLM Agents
202510 papers
- ICRA (T1)Anti-Sensing: Defense against Unauthorized Radar-based Human Vital Sign Sensing with Physically Realizable Wearable Oscillators
- ICLR (T1)Concept Bottleneck Large Language Models
- NAACL (T1)Effective Skill Unlearning through Intervention and Abstention
- ICML (T1)Evaluating Neuron Explanations: A Unified Framework with Sanity Checks
- CVPR (T1)Interpretable Generative Models through Post-hoc Concept Bottlenecks
- TMLR (T1)Interpreting Neurons in Deep Vision Networks with Language Models
- featured@Iterative Self-Tuning LLMs for Enhanced Jailbreaking Capabilities
- WACVSAND: Enhancing Open-Set Neuron Descriptions through Spatial Awareness
- EMNLP (T1)ThinkEdit: Interpretable Weight Editing to Mitigate Overly Short Thinking in Reasoning Models
- ICML (T1)Understanding Fixed Predictions via Confined Regions
202413 papers
- NeurIPS (T1)Abstracted Shapes as Tokens - A Generalizable and Interpretable Model for Time-series Classification
- ICML (T1)AND: Audio Network Dissection for Interpreting Deep Acoustic Models
- ICML (T1)Breaking the Barrier: Enhanced Utility and Robustness in Smoothed DRL Agents
- TMLR (T1)Concept-Driven Continual Learning | [code] | [project page]
- AAAI DAI workshopConcept-Monitor: Understanding DNN training through individual neurons
- ICML Mechanistic Interpretability workshopCrafting Large Language Models for Enhanced Interpretability
- ECCV (T1)Interpretability-Guided Test-Time Adversarial Defense | [code] | [project page]
- ICML (T1)Linear Explanations for Individual Neurons
- ICLR (T1)Prediction without Preclusion: Recourse Verification with Reachable Sets
- NeurIPS (T1)Probabilistic Federated Prompt-Tuning with Non-IID and Imbalanced Data
- NeurIPS (T1)Provable and Efficient Dataset Distillation for Kernel Ridge Regression
- ICLR (T1)Provably Robust Conformal Prediction with Improved Efficiency
20239 papers
- NeurIPS (T1)Analyzing Generalization of Neural Networks through Loss Path Kernels
- ICDMAttacking c-MARL More Effectively: A Data Driven Approach
- ICLR , notable-top-25%CLIP-Dissect: Automatic Description of Neuron Representations in Deep Vision Networks
- ICML (T1)ConCerNet: A Contrastive Learning Based Framework for Automated Conservation Law Discovery and Trustworthy Dynamical System Prediction
- ICCV (T1)Corrupting Neuron Explanations of Deep Visual Features
- ICLR (T1)Label-free Concept Bottleneck Models | [code]
- ICLR (T1)Min-Max Bilevel Multi-objective Optimization with Applications in Robust Machine Learning
- ICDMPromoting Robustness of Randomized Smoothing: Two Cost-Effective Approaches
- NeurIPS ATTRIB workshopThe Importance of Prompt Tuning for Automated Neuron Explanations
Where alumni went4 listed, 100% to industry
Industry 3Academia 0
Most often to
- Amazon
- Apple
- Latitude AI
Apply to UCSDFall 2027
UCSD CSE Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- USUSC
Bistra Dilkina
- Learning theory & optimization
- USUCSD
Parinaz Naghizadeh
- Reasoning & neuro-symbolic AI
- SGNTU
Tianwei Zhang
- Trustworthy & safe ML
- USUT Austin
Kaushik Chowdhury
- Trustworthy & safe ML
- USUCLA
Adnan Darwiche
- Reasoning & neuro-symbolic AI
- GBImperial
Geoffrey Li
- Signal processing & information theory
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.
UCSDcloses