Alexey Kravets

I am a PhD student in Computer Science at the University of Bath, advised by Vinay P. Namboodiri. My research focuses on vision-language models, few-shot learning, machine unlearning, and more recently mechanistic interpretability.

Before my PhD I spent five years as a Lead Data Scientist at Aviva, working on machine learning and NLP projects in the healthcare claims team.

Email / LinkedIn / CV / Google Scholar / GitHub / Medium

Research

I work at the intersection of vision and language. Representative publications are below.

	Interpretability Transfer from Language to Vision via Sparse Autoencoders Alexey Kravets, Da Li, Chuan Li, Da Chen, Vinay P. Namboodiri International Conference on Machine Learning (ICML), 2026 We align textual Sparse Autoencoders with visual representations transferring interpretability to large vision- language models (VLMs) at minimal cost, enabling both interpretability and steering of concepts within VLMs.
	Rethinking Few-Shot CLIP Benchmarks: A Critical Analysis in the Inductive Setting Alexey Kravets, Da Chen, Vinay P. Namboodiri International Conference on Computer Vision (ICCV), 2025 We identify a flaw in the evaluation of existing few-shot methods with CLIP and propose a pipeline to evaluate them more fairly using unlearning.
	Zero-shot CLIP Class Forgetting via Text-Image Space Adaptation Alexey Kravets, Vinay P. Namboodiri Transactions on Machine Learning Research (TMLR), 2025 We propose a class removal technique for CLIP without using any images, but only text of the class to forget. The method relies on changing the text projection matrix in CLIP.
	Zero-Shot Class Unlearning in CLIP with Synthetic Samples Alexey Kravets, Vinay P. Namboodiri Winter Conference on Applications of Computer Vision (WACV), 2025 We unlearn classes from CLIP without real data by generating synthetic samples and applying Lipschitz regularization.
	Addressing Blind Guessing: Calibration of Selection Bias in Multiple-Choice Question Answering by Video Language Models Olga Loginova, Oleksandr Bezrukov, Ravi Shekhar, Alexey Kravets Annual Meeting of the Association for Computational Linguistics (ACL), 2025 We show that video-language models suffer from positional bias in multiple-choice QA and introduce a post-processing calibration technique based on fairness bias metrics.
	CLIP Adaptation by Intra-modal Overlap Reduction (Oral) Alexey Kravets, Vinay P. Namboodiri British Machine Vision Conference (BMVC), 2024 We improve training-free few-shot learning methods that use CLIP by reducing the intra-modal overlap in the CLIP image encoder.

Template design from Jon Barron.