Ivaxi Sheth

I am a PhD student at CISPA Helmholtz Center, supervised by Prof. Mario Fritz. I hold a BEng + MEng (Hons) from Imperial College London, where I was supervised by Dr. Carlo Ciliberto. Previously, I was a Research Assistant at Mila – Quebec AI Institute, and an AI Research Engineer at Imagination Technologies (UK), working under Dr. Cagatay Dikici on hardware acceleration for neural networks.

My research currently revolves around two main themes:

  1. Causal reasoning and scientific agents, with a focus on how multi-agent LLM systems can assist in discovering, refining, and validating causal hypotheses.
  2. Safety and ethics of multi-agent and self-evolving AI systems, with an emphasis on long-term risk from autonomous, open-ended and memory-augmented agents.

More recently, I have been investigating persistent memory for LLMs, mechanisms that allow models to retain information across sessions and internal states. While such memory can enable personalization and long-horizon reasoning, it also introduces significant safety risks, including hidden state accumulation, privacy leakage, and long-term behavioral manipulation.

I enjoy learning from and collaborating with people from diverse backgrounds and disciplines. My inbox is always open, feel free to reach out.

News
  • Jan 2026: Funny or Persuasive, but Not Both: Evaluating Fine-Grained Multi-Concept Control in LLMs accepted at EACL 2026 🇲🇦 !
  • Dec 2025: Survey on AI Ethics: A Socio‐Technical Perspective accepted!
  • Nov 2025: Context-Aware Reasoning On Parametric Knowledge for Inferring Causal Variables presented at EMNLP in 🇨🇳!
  • Aug 2025: Started Applied Scientist II Internship at Amazon Science.
  • May 2025: Presenting at NAACL 2025 in New Mexico.
Publications
Beyond Steering thumbnail
Funny or Persuasive, but Not Both: Evaluating Fine-Grained Multi-Concept Control in LLMs
Arya Labroo, Ivaxi Sheth, Vyas Raina, Amaani Ahmed, Mario Fritz.
EACL Main 2026
PersistBench thumbnail
PersistBench: When Should Long-Term Memories Be Forgotten by LLMs?
Sidharth Pulipaka, Oliver Chen, Manas Sharma, Taaha Bajwa, Vyas Raina, Ivaxi Sheth.
Preprint

Benchmark long-term memory usage in LLMs and find that persistent memories systematically induce cross-domain leakage and sycophantic behavior.

Open-Ended AI Safety thumbnail
Safety Must Precede the Deployment of Open-Ended AI Agents
Ivaxi Sheth, Jan Wehner, Sahar Abdelnabi, Ruta Binkyte, Mario Fritz.
Preprint · Earlier version accepted at ICLR SSI-FM 2025

Examines systemic risks in open-ended, self-propagating AI agents and proposes mitigation strategies.

Invariance Conflicts thumbnail
Trustworthy AI Suffers from Invariance Conflicts and Causality is the Solution
Ruta Binkyte*, Ivaxi Sheth*, Zhijing Jin, Mohammad Havaei, Bernhard Schölkopf, Mario Fritz.
Preprint

Frames trustworthy AI trade-offs as incompatible invariance requirements and positions causality as a unifying solution.

Justice in Judgment thumbnail
Justice in Judgment: Unveiling (Hidden) Bias in LLM-Assisted Peer Reviews
Sai Suresh Macharla Vasu*, Ivaxi Sheth*, Hui-Po Wang, Ruta Binkyte, Mario Fritz.
ARR

LLM-assisted peer reviews exhibit biases against certain author demographics and academic expertise levels.

IV Co-Scientist thumbnail
IV Co-Scientist: Multi-Agent LLM Framework for Instrumental Variable Discovery
Ivaxi Sheth, Zhijing Jin, Bryan Wilder, Dominik Janzing, Mario Fritz.
ARR

Multi-agent LLM framework for discovering instrumental variables from large observational datasets and databases.

AI Ethics Survey thumbnail
Survey on AI Ethics: A Socio-technical Perspective
Dave Mbiazi, Meghana Bhange, Maryam Babaei, Ivaxi Sheth, Patrik Joslin Kenfack, Samira Ebrahimi Kahou.
Wiley Computational Intelligence · 2025

Comprehensive survey of AI ethics from a socio-technical perspective, covering key challenges.

Context-Aware Reasoning thumbnail
Context-Aware Reasoning on Parametric Knowledge for Inferring Causal Variables
Ivaxi Sheth, Sahar Abdelnabi, Mario Fritz.
EMNLP Findings 2025

LLM framework for hypothesizing causal variables and building a causal graph from parametric knowledge and context.

CausalGraph2LLM thumbnail
CausalGraph2LLM: Evaluating LLMs for Causal Queries
Ivaxi Sheth, Bahare Fatemi, Mario Fritz.
NAACL Findings 2025

Benchmark for causal-graph reasoning and sensitivity to graph encoding in LLMs.

LLM Task Interference thumbnail
LLM Task Interference: An Initial Study on the Impact of Task-Switch in Conversational History
Akash Gupta*, Ivaxi Sheth*, Vyas Raina, Mark Gales, Mario Fritz.
EMNLP Main 2024

Studies performance degradation when models must switch tasks for multi-turn conversations.

Concept-based models thumbnail
Auxiliary Losses for Learning Generalizable Concept-based Models
Ivaxi Sheth, Samira Ebrahimi Kahou.
NeurIPS 2023

Adds inductive bias via auxiliary objectives to improve generalization in concept bottleneck models.

LLM4GRN thumbnail
LLM4GRN: Discovering Causal Gene Regulatory Networks with LLMs -- Evaluation through Synthetic Data Generation
Tejumade Afonja*, Ivaxi Sheth*, Ruta Binkyte*, Waqar Hanif, Thomas Ulas, Matthias Becker, Mario Fritz.
Preprint

Explores LLM-assisted pipelines for inferring gene regulatory networks from RNA sequencing data.

MedG-KRP: Medical Graph Knowledge Representation Probing
Gabriel R. Rosenbaum, Lavender Yao Jiang*, Ivaxi Sheth*, Eric Karl Oermann.
ML4H

Benchmark for evaluating LLMs' medical knowledge using graph-based probing tasks.

Transparent Anomaly Detection thumbnail
Transparent Anomaly Detection via Concept-based Explanations
Laya Rafiee*, Ivaxi Sheth*, Farhood Farahnak*, Shirin Abbasinejad Enger.
XAI in Action: Past, Present, and Future Applications · NeurIPS 2023

Introduces concept-based explanations for anomaly detection models.

Uncertain Concepts thumbnail
Learning from Uncertain Concepts via Test-Time Interventions
Ivaxi Sheth, Aamer Abdul, Laya Rafiee, Mohammad Havaei, Samira Ebrahimi Kahou.
Trustworthy and Socially Responsible Machine Learning · NeurIPS 2022

Proposes a framework for handling uncertain concepts in concept-based models using test-time interventions.

FHIST: A Benchmark for Few-Shot Classification of Histological Images
Fereshteh Shakeri, Malik Boudiaf, Sina Mohammadi, Ivaxi Sheth et al.
Preprint

Realistic clinical benchmark for few-shot learning in histology.

Three-Stream Network for Enriched Action Recognition
Ivaxi Sheth.
CVPR Workshops 2021
Mentees
Akash Gupta University of Cambridge → Research Engineer at Apta
Arya Labroo University of Cambridge
Gabriel R. Rosenbaum University of Chicago
Sai Suresh Macharla Vasu CISPA
Sidharth Pulipaka SPAR
Oliver Chen SPAR
Taaha Bajwa SPAR
Manas Sharma SPAR
Services

Reviewer: ARR, NeurIPS, ICML, ICLR
Workshop Organizer: Women in Computer Vision at CVPR 2022 and CVPR 2023

Patents
Hardware implementation of windowed operations in three or more dimensions
Ivaxi Sheth, Cagatay Dickici, Aria Ahamdi, James Imber
Patent · US Patent 12,198,034
System and method of performing convolution efficiently adapting Winograd algorithm
Ferdinando Insalata, Cagatay Dikici, James Imber, Ivaxi Sheth
Patent · US Patent Application 18/613,443
Misc.

I am a trained Kathak dancer (Indian classical dance), and I regularly teach and perform. In the summers, I enjoy container gardening. I also like to travel and cook veggie food 🍲. More recently, I have taken up punch-needle art 🌸.