Publications.
Papers and preprints on learning dynamics, interpretability, and applications of machine learning.
Google Scholar ↗9 publications
Navigating Potholes with Geometry-Aware Sharpness Minimization
Combining a slowly learned loss geometry with sharpness-aware perturbations to improve neural network optimization.
T-GRAB: A Synthetic Diagnostic Benchmark for Learning on Temporal Graphs
Controlled synthetic tasks that test temporal graph models on repetition, delayed causal effects, and long-range dependencies.
Earlier version presented at the MLoG-GenAI Workshop at KDD 2025 (oral).
Interpretability in Action: Exploratory Analysis of VPT, a Minecraft Agent
Attention analysis and interventions in a Minecraft agent reveal clues about memory and a case of goal misgeneralization.
Bridging the Neural Divide: Geometrical Insights and Practical Approaches to Machine Learning Interpretability
A study of neural network geometry, reverse engineering, and practical interpretability in a vision-based agent.
Reverse Engineering Deep ReLU Networks An Optimization-based Algorithm
An optimization-based approach to reconstructing deep ReLU networks from input-output queries.
Predicting survival of Iranian COVID-19 patients infected by various variants including omicron from CT Scan images and clinical data using deep neural networks
Combining clinical information and chest CT images to study patient survival prediction.
A New R Package for Categorizing Coding and Non-Coding Genes
An R package for categorizing coding and non-coding genes.
Somatic point mutations are enriched in non-coding RNAs with possible regulatory function in breast cancer
Investigating somatic mutations in non-coding RNAs and their potential regulatory roles in breast cancer.
Accurate and Rapid Diagnosis of COVID-19 Pneumonia with Batch Effect Removal of Chest CT-Scans and Interpretable Artificial Intelligence
Interpretable deep learning for chest CT analysis, with preprocessing to reduce differences across medical centers.
No publications match. Try another search or choose “All”.