Interview with Deepika Vemuri: interpretability and concept-based learning

2026-04-24 · Source: ΑΙhub · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, medium

Summary

Deepika Vemuri, a PhD student at IIT Hyderabad, is researching interpretability and concept-based learning in AI models, particularly for high-stakes applications like criminal justice and healthcare. Her work aims to guide model learning through human-understandable concepts, similar to how humans learn by identifying shared features. She has explored two main aspects: enhancing concept-to-class mapping beyond linear layers by introducing a logic module with differentiable fuzzy logic gates, which improved accuracy and interpretability, leading to a WACV 2026 publication. Additionally, she investigated aligning neural network depth-wise representations with semantic hierarchies, guiding models to learn general concepts in early layers and specific ones in deeper layers using Formal Concept Analysis, resulting in more interpretable embeddings and hierarchically structured concept representations. She is also exploring concept representation in videos and developing new metrics for concept-based models.

Key takeaway

For research scientists and computer vision engineers developing interpretable AI, consider integrating logic modules or hierarchical concept learning into your models. This approach can improve both model accuracy and the clarity of concept-to-class associations, which is vital for deploying AI in sensitive domains. You should also explore new evaluation metrics beyond task accuracy to better assess the robustness and reliability of concept-based models against misleading information.

Key insights

Concept-based learning enhances AI interpretability by aligning model representations with human-understandable concepts.

Principles

• Interpretability is crucial for high-stakes AI applications.
• Logic operations can model complex concept-to-class relations.
• Semantic hierarchies improve concept learning in deep networks.

Method

A logic module with differentiable fuzzy logic gates learns predicates for classes. Formal Concept Analysis constructs a concept lattice to provide supervisory signals for hierarchical concept learning across network layers.

In practice

• Use fuzzy logic gates for non-linear concept associations.
• Structure concept learning hierarchically across network layers.
• Develop worst-case metrics for concept-based model evaluation.

Topics

• Interpretability
• Concept-Based Learning
• Logic Modules
• Formal Concept Analysis
• Video Language Models

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, AI Student

Related on AIssential

• Improving AI models’ ability to explain their predictions — Extracting inherent model concepts yields more accurate and interpretable AI explanations than predefined human concepts.
• Interpretability and implicit model semantics in biomedicine and deep learning — Interpretability is a single aspect of a deep learning model's overall semantics, informed by philosophy of science.
• Towards Intrinsic Interpretability of Large Language Models:A Survey of Design Principles and Architectures — Intrinsic interpretability builds transparency directly into LLM architectures to enhance trustworthiness and safety.
• “Features” aren’t always the true computational primitives of a model, but that might be fine anyways — Features in mechanistic interpretability exist on a spectrum from memorization to computational primitives.
• Translating black-box medical AI models into interpretable global decision logic — Class-Association Manifold Learning (CAML) explains black-box medical AI by visualizing global decision rules on low-dimensional manifolds.
• Enhancing AI Interpretability and Safety through Localised Architectures — Localized hardware ML architectures can enhance AI interpretability and efficiency by leveraging higher per-node expressivity, particularly for smaller datasets.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ΑΙhub.