Akash Pandey

Akash Pandey

PhD Candidate, Northwestern University

Akash Pandey

I have a Ph.D. from Northwestern University, where I worked on explainable deep learning for scientific discovery. I was co-advised by Dr. Sinan Keten and Dr. Wei Chen. Prior to this, I completed a Master’s by research in Applied Mechanics at IIT Madras, specializing in Applied mechanics. My industry experience includes two internships at Capital One (recommendation systems and eXplainable AI) and a full-time position at Rolls Royce.

My research develops interpretable and generative AI models at the intersection of computational biology, time-series analysis, and recommendation systems. A central theme across my work is building explainability directly into the learning pipeline, rather than treating it as a post-hoc afterthought.

On the explainability side, I developed COLOR and TimeSliver, novel CNN-based architectures that improve monomer-level interpretability for biological sequences by 22% and temporal explainability for time-series classification by 11%, while maintaining Transformer-level predictability (ICLR ‘26, ICLR ‘25 MLGenX). I have also published in Nature Communications Materials, ACS JCIM, and Cell Press Patterns.

On the generative modeling side, I integrate attribution signals into the design loop. I developed an attribution-guided evolutionary learning framework for black-box optimization of biological sequences, improving sample efficiency by 19% (ICML ‘26, ICLR ‘26 MLGenX). My most recent work develops explainability-steered VAE and diffusion models for protein and DNA design, achieving a 62% improvement in sample efficiency (under submission at NeurIPS ‘26).

Beyond biology, I enjoy applying deep learning to other sequential modalities. I designed a LLaMA-3 based model for EMG-to-text conversion, reducing word error rate by 20% over baselines (ACL ‘25 Main). I have also published work on audio-based emotion detection (ACM Multimedia ‘23) and biosignal-based person identification (ICASSP ‘23).

You can reach me at akash.pandey@northwestern.edu
Link to my CV

Recent Highlights

  • [May, 2026] - Paper on explainability-steered VAE and diffusion frameworks for biological sequence design submitted to NeurIPS ‘26.
  • [Jan 2026] - Paper on attribution-guided evolutionary learning for biological sequence optimization accepted at ICML ‘26 and ICLR ‘26 MLGenX.
  • [Nov, 2026] - TimeSliver accepted at ICLR ‘26.
  • [March, 2025] - Paper on surface-EMG based silent speech recognition using LLaMA-3 accepted at ACL ‘25 Main.
  • [March, 2025] - COLOR accepted at MLGenX workshop, ICLR 2025.
  • [June-August, 2025] - Completed Data Science PhD internship at Capital One, San Jose.

Services

  • Guiding Julia Levenshteyn, a Senior at Lake Forest College, to develop metrics for quantifying explainability in computational biology.
  • Guided Xiaoyuan Zhang, a Master’s student in EECS at Northwestern University, to develop a LLaMA-based deep learning model for silent EMG-to-text prediction on a closed vocabulary.
  • Guided Yueyuan Sui, a Master’s student in EECS at Northwestern University, for the ACM MM 2023 challenge on human emotion prediction.
  • Reviewer at Nature Computational Materials and IEEE/ACM Transactions on Computational Biology and Bioinformatics.