Haris Vikalo – Selected projects

Haris Vikalo – Selected Projects

Interpretable AI for Genomic Perturbation Response

This project studies multimodal and interpretable machine learning methods for predicting how genomic perturbations affect cellular behavior across contexts. The emphasis is on combining pre-perturbation cellular state, chromatin context, and perturbation information to support both accurate prediction and biologically meaningful interpretation.

Ongoing directions include benchmark design, multimodal perturbation modeling, and structured latent representations that can be linked to reusable regulatory programs. This effort builds naturally on prior work adapting genomic foundation models to difficult biological prediction tasks.

  • Working project title: OpenPerturb
  • Related prior work: XVir, NextVir
  • Current emphasis: open benchmarks, interpretable multimodal models, and reusable scientific infrastructure
Learn more →

Efficient and Adaptive AI at the Edge

This project studies inference and learning systems that operate under realistic resource constraints, including limited communication bandwidth, heterogeneous hardware, privacy constraints, and nonstationary environments. The goal is to make AI systems more efficient, robust, and adaptive in edge and distributed settings.

Representative themes include distributed inference orchestration, scalable learning over constrained networks, adaptive model management, and continual or federated learning in dynamic environments. The focus is on methods that remain useful when theory, algorithms, and systems considerations must all be addressed together.

  • Representative themes: edge inference, federated learning, continual adaptation, and resource-aware AI systems
  • Current emphasis: efficient deployment, robustness to change, and practical system-aware learning algorithms
Learn more →