Aditya Taparia

PhD Student, Computer Science · Arizona State University · LENS Lab

Hi! I'm Aditya Taparia, a second-year Computer Science PhD student at Arizona State University, advised by Prof. Ransalu Senanayake and affiliated with the Laboratory for Learning Evaluation and Naturalization of Systems (LENS Lab). My research is centered on the conceptual foundations of neural networks—exploring how models form, refine, and leverage high-level abstractions or “concepts” to learn effectively and make informed decisions. I aim to develop concept-driven AI systems that are more robust, adaptable, and capable of handling distribution shifts in evolving real-world environments. By grounding learning and decision-making in explicit concepts, I seek to improve model interpretability, generalization, and optimize AI-driven decision processes.

Most recently I was a Research Intern at Bosch (2026), working on explainability, and before that a Student Research Intern in the MIG group at Lawrence Livermore National Laboratory (2025), where I worked on uncertainty quantification for large language models.

Previously, I completed my Bachelors with Honors in Computer Science from the Indian Institute of Information Technology (IIIT), Kottayam.

I'm always happy to connect about research, collaborate on ideas, or share advice. Please feel free to get in touch!

Aditya Taparia
email Email picture_as_pdf CV school Google Scholar LinkedIn GitHub

News

May '26Our paper "VLC Fusion: Vision-Language Conditioned Sensor Fusion for Robust Object Detection" was accepted to ECML PKDD'26 and shortlisted for the Best Research Track Student Paper Award 🎗.
May '26Joined Bosch as a Research Intern.
Jan '26Our paper "ExpressivityBench: Can LLMs Communicate Implicitly?" was accepted to Findings of EACL'26.
Nov '25Our paper "Consistency-based Abductive Reasoning over Perceptual Errors of Multiple Pre-trained Models in Novel Environments" was accepted to AAAI'26.
Jun '25Our paper "BaTCAVe: Trustworthy Explanations for Robot Behaviors" was accepted to IROS'25.
May '25Joined Lawrence Livermore National Laboratory as a Student Research Intern in the MIG group.
May '25Our paper "Explainable Concept Generation through Vision-Language Preference Learning for Understanding Neural Networks' Internal Representations" was accepted to ICML'25.
Jan '25Joined Arizona State University as a PhD student in Computer Science.
Dec '24Presented four workshop papers at NeurIPS'24.
May '24Our paper "Failures Are Fated, But Can Be Faded: Characterizing and Mitigating Unwanted Behaviors in Large-Scale Vision and Language Models" was accepted to ICML'24 as a spotlight-designated paper 🎗 (top 3.5%).
Aug '23Joined Arizona State University as a Masters student in Computer Science.

Research

See my CV and Google Scholar.

All (11) Conferences & Journals (8) Preprints & Workshops (3)
  1. VLC Fusion architecture: LiDAR and camera features are combined with vision-language model conditions through a CBAM, BatchNorm and FiLM block, raising detections from 6/20 to 12/20 objects compared with standard multi-modal fusion.

    VLC Fusion: Vision-Language Conditioned Sensor Fusion for Robust Object Detection

    Aditya Taparia, Noel Ngu, Mario Leiva, Joshua Shay Kricheli, John Corcoran, Nathaniel D. Bastian, Gerardo Simari, Paulo Shakarian, Ransalu Senanayake
    European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD), 2026 · Shortlisted for Best Research Track Student Paper Award 🎗
    A sensor fusion framework that uses a vision-language model to read environmental context and adaptively reweight modalities, making object detection robust to lighting, weather, and occlusion shifts.
    picture_as_pdf PDFarticle Details
    @inproceedings{taparia2026vlc,
    title={VLC Fusion: Vision-Language Conditioned Sensor Fusion for Robust Object Detection},
    author={Taparia, Aditya and Ngu, Noel and Leiva, Mario and Kricheli, Joshua Shay and Corcoran, John and Bastian, Nathaniel D. and Simari, Gerardo and Shakarian, Paulo and Senanayake, Ransalu},
    booktitle={European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD)},
    year={2026}
}

  2. The Anatomy of Uncertainty in LLMs

    Aditya Taparia, Ransalu Senanayake, Kowshik Thopalli, Vivek Sivaraman Narayanaswamy
    ICBINB Workshop at the International Conference on Learning Representations (ICLR), 2026
    Decomposes LLM uncertainty into input ambiguity, knowledge gaps, and decoding randomness, showing their relative weight changes with model scale and task — a more actionable signal than a single confidence score.
    picture_as_pdf PDFarticle Details
    @inproceedings{taparia2026anatomy,
    title={The Anatomy of Uncertainty in LLMs},
    author={Taparia, Aditya and Senanayake, Ransalu and Thopalli, Kowshik and Narayanaswamy, Vivek Sivaraman},
    booktitle={ICBINB Workshop at the International Conference on Learning Representations (ICLR)},
    year={2026}
}
  3. Overview figure for ExpressivityBench, showing the generator-to-grader channel used to measure how much implicit signal a language model transmits.

    ExpressivityBench: Can LLMs Communicate Implicitly?

    Joshua Tint, Som Sagar, Aditya Taparia, Kelly Raines, Bimsara Pathiraja, Caleb Liu, Ransalu Senanayake
    Findings of the Association for Computational Linguistics: EACL, 2026
    An information-theoretic benchmark for implicit communication in LLMs across nine tasks, showing models convey emotion and tone well but lag on sociolinguistic signals like age, gender, and political slant.
    picture_as_pdf PDFarticle Details
    @inproceedings{tint2026expressivitybench,
    title={ExpressivityBench: Can LLMs Communicate Implicitly?},
    author={Tint, Joshua and Sagar, Som and Taparia, Aditya and Raines, Kelly and Pathiraja, Bimsara and Liu, Caleb and Senanayake, Ransalu},
    booktitle={Findings of the Association for Computational Linguistics: EACL 2026},
    pages={4500--4515},
    year={2026}
}

  4. Learning to Configure Agentic AI Systems

    Aditya Taparia, Som Sagar, Ransalu Senanayake
    arXiv preprint, 2026
    ARC treats agent configuration as a semi-Markov decision process and learns to pick a query-specific configuration per request, improving reasoning accuracy by 31.3% and tool-use accuracy by 13.95% over fixed pipelines.
    picture_as_pdf PDFarticle Details
    @article{taparia2026learning,
    title={Learning to Configure Agentic AI Systems},
    author={Taparia, Aditya and Sagar, Som and Senanayake, Ransalu},
    journal={arXiv preprint arXiv:2602.11574},
    year={2026}
}

  5. Consistency-based Abductive Reasoning over Perceptual Errors of Multiple Pre-trained Models in Novel Environments

    Mario A. Leiva, Noel Ngu, Joshua Shay Kricheli, Aditya Taparia, Ransalu Senanayake, Paulo Shakarian, Nathaniel D. Bastian, John Corcoran, Gerardo I. Simari
    Proceedings of the AAAI Conference on Artificial Intelligence, 2026
    Reconciles conflicting predictions from multiple pre-trained perception models under distribution shift by treating it as a test-time consistency-based abduction problem, beating the best single model by ~13.6% F1 across 15 datasets.
    picture_as_pdf PDFarticle Details
    @inproceedings{leiva2026consistency,
    title={Consistency-based Abductive Reasoning over Perceptual Errors of Multiple Pre-trained Models in Novel Environments},
    author={Leiva, Mario A. and Ngu, Noel and Kricheli, Joshua Shay and Taparia, Aditya and Senanayake, Ransalu and Shakarian, Paulo and Bastian, Nathaniel D. and Corcoran, John and Simari, Gerardo I.},
    booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
    pages={19216--19223},
    year={2026}
}
  6. Overview figure for BaTCAVe, showing Bayesian concept activation vectors producing explanations of robot decisions with uncertainty estimates.

    BaTCAVe: Trustworthy Explanations for Robot Behaviors

    Som Sagar, Aditya Taparia, Harsh Mankodiya, Pranav Bidare, Yifan Zhou, Ransalu Senanayake
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025
    A Bayesian concept-based explainability method that reports calibrated uncertainty alongside each explanation of a robot's decision, validated in simulation and on real robots.
    picture_as_pdf PDFplay_circle Videocode Codearticle Details
    @inproceedings{sagar2025batcave,
    title={BaTCAVe: Trustworthy Explanations for Robot Behaviors},
    author={Sagar, Som and Taparia, Aditya and Mankodiya, Harsh and Bidare, Pranav and Zhou, Yifan and Senanayake, Ransalu},
    booktitle={2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
    pages={13867--13874},
    year={2025}
}
  7. Overview figure for RLPO, showing a diffusion model fine-tuned by reinforcement learning preference optimization to generate concept images that explain a classifier.

    Explainable Concept Generation through Vision-Language Preference Learning for Understanding Neural Networks' Internal Representations

    Aditya Taparia, Som Sagar, Ransalu Senanayake
    International Conference on Machine Learning (ICML), 2025
    Frames concept-based explanation as image generation and uses RL preference optimization to automatically discover what concepts a network has internally learned, including ones humans would not think to name.
    picture_as_pdf PDFplay_circle Videocode Codearticle Details
    @inproceedings{taparia2025explainable,
    title={Explainable Concept Generation through Vision-Language Preference Learning for Understanding Neural Networks' Internal Representations},
    author={Taparia, Aditya and Sagar, Som and Senanayake, Ransalu},
    booktitle={Proceedings of the 42nd International Conference on Machine Learning},
    year={2025}
}
  8. Sample imagery from the MDS-A dataset, showing aerial scenes rendered under multiple simulated distribution shifts.

    Multiple Distribution Shift-Aerial (MDS-A): A Dataset for Test-Time Error Detection and Model Adaptation

    Noel Ngu, Aditya Taparia, Gerardo I. Simari, Mario A. Leiva, Ransalu Senanayake, Paulo Shakarian, Nathaniel D. Bastian, John Corcoran
    Proceedings of the AAAI Symposium Series, 2025
    An aerial imagery dataset with explicitly controlled distribution shifts, designed for benchmarking test-time error detection and model adaptation.
    picture_as_pdf PDFarticle Details
    @inproceedings{ngu2025multiple,
    title={Multiple Distribution Shift-Aerial (MDS-A): A Dataset for Test-Time Error Detection and Model Adaptation},
    author={Ngu, Noel and Taparia, Aditya and Simari, Gerardo I. and Leiva, Mario A. and Senanayake, Ransalu and Shakarian, Paulo and Bastian, Nathaniel D. and Corcoran, John},
    booktitle={Proceedings of the AAAI Symposium Series},
    volume={5},
    pages={379--383},
    year={2025}
}
  9. Overview figure for the red teaming work, showing a language model scoring mismatches between prompts and diffusion model outputs to produce a reward signal.

    LLM-Assisted Red Teaming of Diffusion Models through "Failures Are Fated, But Can Be Faded"

    Som Sagar, Aditya Taparia, Ransalu Senanayake
    NeurIPS Workshop on Red Teaming GenAI, 2024
    Extends the Failures Are Fated framework to text-to-image diffusion models, using an LLM to generate reward signals and states for reinforcement learning-driven red teaming.
    picture_as_pdf PDFarticle Details
    @inproceedings{sagar2024llm,
    title={LLM-Assisted Red Teaming of Diffusion Models through "Failures Are Fated, But Can Be Faded"},
    author={Sagar, Som and Taparia, Aditya and Senanayake, Ransalu},
    booktitle={Red Teaming GenAI: What Can We Learn from Adversaries?},
    year={2024}
}

  10. Transforming satellite imagery into vector maps using modified GANs

    Aditya Taparia, Ali Kashif Bashir, Yaodong Zhu, Thippa Reddy Gdekallu, Keshab Nath
    Alexandria Engineering Journal, 2024
    HPix uses a hierarchical modified GAN to generate vector tile maps directly from satellite images, for building footprints, disaster analysis, and urban planning.
    picture_as_pdf PDFarticle Details
    @article{taparia2024transforming,
    title={Transforming satellite imagery into vector maps using modified GANs},
    author={Taparia, Aditya and Bashir, Ali Kashif and Zhu, Yaodong and Gdekallu, Thippa Reddy and Nath, Keshab},
    journal={Alexandria Engineering Journal},
    volume={109},
    pages={792--806},
    year={2024},
    doi={10.1016/j.aej.2024.09.074}
}
  11. Overview figure for Failures Are Fated, showing a deep reinforcement learning agent exploring and reshaping the failure landscape of a large vision and language model.

    Failures Are Fated, But Can Be Faded: Characterizing and Mitigating Unwanted Behaviors in Large-Scale Vision and Language Models

    Som Sagar, Aditya Taparia, Ransalu Senanayake
    International Conference on Machine Learning (ICML), 2024 · Spotlight (top 3.5%) 🎗
    A deep RL method that systematically explores where large vision-language models fail, then uses limited human feedback to reshape that failure landscape and mitigate bias and misalignment.
    picture_as_pdf PDFplay_circle Videocode Codearticle Details
    @inproceedings{sagar2024failures,
    title={Failures are fated, but can be faded: characterizing and mitigating unwanted behaviors in large-scale vision and language models},
    author={Sagar, Som and Taparia, Aditya and Senanayake, Ransalu},
    booktitle={Proceedings of the 41st International Conference on Machine Learning},
    pages={42999--43023},
    year={2024}
}