Anjishnu Mukherjee

2026 2 publications

ArXiv

KnowBias: Mitigating Social Bias in LLMs via Know-Bias Neuron Enhancement

Jinhao Pan, Chahat Raj, Anjishnu Mukherjee, Sina Mansouri, Bowen Wei, Shloka Yada, Ziwei Zhu.

Paper Code

Highlights and amplifies neurons tied to bias-related knowledge so a model can preserve capability while reducing stereotyped generations across social groups.

ArXiv

Metadata Conditioned Large Language Models for Localization

Anjishnu Mukherjee, Ziwei Zhu, Antonios Anastasopoulos.

Paper Code

Conditions LLM behavior on structured cultural and regional metadata to improve localization quality without requiring a separate model for every setting.

2025 2 publications

ArXiv

Measuring South Asian Biases in Large Language Models

Mamnuya Rinki, Chahat Raj, Anjishnu Mukherjee, and Ziwei Zhu.

Paper

Benchmarks how large language models portray South Asian identities, surfacing recurring stereotypes, representational gaps, and uneven treatment across prompts and groups.

WACV '25

Crossroads of Continents: Automated Artifact Extraction for Cultural Adaptation with Large Multimodal Models

Anjishnu Mukherjee, Ziwei Zhu, Antonios Anastasopoulos.

Paper Code

Uses multimodal models to extract culturally salient artifacts from images so visual content can be adapted more appropriately across regional contexts.

2024 4 publications

EMNLP Findings '24

BiasDora: Exploring Hidden Biased Associations in Vision-Language Models

Chahat Raj, Anjishnu Mukherjee, Aylin Caliskan, Antonios Anastasopoulos, Ziwei Zhu.

Paper Code

Probes vision-language models for hidden biased associations, showing how multimodal representations can encode and reproduce social stereotypes.

AIES '24

Breaking Bias, Building Bridges: Evaluation and Mitigation of Social Biases in LLMs via Contact Hypothesis

Chahat Raj*, Anjishnu Mukherjee*, Aylin Caliskan, Antonios Anastasopoulos, Ziwei Zhu.

Paper Code

Evaluates whether contact-hypothesis-inspired interventions can reduce social bias in LLM outputs while maintaining the model’s overall usefulness.

NAACL '24

Global Gallery: The Fine Art of Painting Culture Portraits through Multilingual Instruction Tuning

Anjishnu Mukherjee, Aylin Caliskan, Ziwei Zhu, Antonios Anastasopoulos.

Paper Code

Builds multilingual instruction-tuning methods that help language models generate culturally grounded depictions instead of flattening diverse visual traditions.

ECIR '24 Random

SALSA: Salience-Based Switching Attack for Adversarial Perturbations in Fake News Detection Models

Chahat Raj*, Anjishnu Mukherjee*, Hemant Purohit, Antonios Anastasopoulos, Ziwei Zhu.

Paper Code

Introduces a salience-based adversarial attack that exposes brittleness in fake-news detectors by perturbing the most influential tokens.

2023 2 publications

EMNLP '23

Global Voices, Local Biases: Socio-cultural Prejudices across Languages

Anjishnu Mukherjee*, Chahat Raj*, Ziwei Zhu, Antonios Anastasopoulos.

Paper Code

Studies how socio-cultural prejudices vary across languages, showing that multilingual systems do not simply inherit one uniform pattern of bias.

AIES '23

True and Fair: Robust and Unbiased Fake News Detection via Interpretable Machine Learning

Chahat Raj, Anjishnu Mukherjee and Ziwei Zhu.

Paper Code

Combines robust modeling with interpretable signals to improve fake-news detection while reducing unwanted bias in the system’s decisions.

Hi, I'm Anjishnu!

Selected Publications

KnowBias: Mitigating Social Bias in LLMs via Know-Bias Neuron Enhancement

Metadata Conditioned Large Language Models for Localization

Measuring South Asian Biases in Large Language Models

Crossroads of Continents: Automated Artifact Extraction for Cultural Adaptation with Large Multimodal Models

BiasDora: Exploring Hidden Biased Associations in Vision-Language Models

Breaking Bias, Building Bridges: Evaluation and Mitigation of Social Biases in LLMs via Contact Hypothesis

Global Gallery: The Fine Art of Painting Culture Portraits through Multilingual Instruction Tuning

SALSA: Salience-Based Switching Attack for Adversarial Perturbations in Fake News Detection Models

Global Voices, Local Biases: Socio-cultural Prejudices across Languages

True and Fair: Robust and Unbiased Fake News Detection via Interpretable Machine Learning