~/meghana/home.qmd

−▢×

> Hello, world. I am —

Meghana Bhange

Researcher ML Engineer PhD @ ÉTS Montréal · Mila

I work at the intersection of machine learning, privacy, and social impact, where my research focuses on Algorithmic Collective Action (ACA) and Protective Optimisation Technologies (POTs) to allow communities to resist algorithmic harm and regain agency in these systems.

Email Mastodon Scholar Toptal GitHub 📍 Montréal, CA

About / Research

scroll ▼

My current research investigates Algorithmic Collective Action (ACA) (Hardt et al., 2023) and Protective Optimisation Technologies (POTs) (Kulynych et al., 2020) to understand how communities can feel empowered to resist harm that occur in algorithmic system and reclaim agency. The main research objective of this work is to investigate POTs (external user-driven interventions) as mechanisms that allow individuals and communities to contest harmful model decisions, opt out of algorithmic systems, modify the algorithmic decision, and get recourse in algorithmic decision-making systems, with the aim of mitigating algorithmic harms.

I pursue this direction as a PhD researcher under the co-supervision of Prof. Ulrich Aïvodji and Prof. Elliot Creager. I am currently affiliated with the Trustworthy Information Systems Lab (TISL) and Mila — Québec AI Institute as a Student Researcher.

I completed my Master’s in Information Technology Engineering at ÉTS Montréal. My thesis focused on building privacy-preserving infrastructure for collective complaint systems, co-supervised by Prof. Aïvodji and Prof. Jean-Marc Robert.

Previously, I worked as a Machine Learning Engineer and still consult via Toptal, contributing to NLP-driven projects like inclusive hiring pipelines (SpaCy, GPT-3, FastAPI) and an automated trading platform integrating technical-indicator strategies.

Motivation for the work

Algorithmic systems are routinely used to make high-stake decisions in a lot of different areas. As these systems are situated in the sociotechnical intersection, they are also prone to introduce harms for users interacting with these systems. Algorithmic harms (Shelby et al., 2023) can occur in various ways in AI driven systems such as perpetuating unjust hierarchies or even lead do uneven distribution of resources. Hence, this creates need for techniques and methods to allow users to mitigate these harms. While algorithmic fairness techniques have been widely proposed, traditional algorithmic fairness techniques assume that platforms are responsible for ensuring that the models are fair, while users receive these fair outcomes or explanations that systems provide. This provider-centric model of accountability limits attention to harms that can be fixed within existing optimization systems and assumes the cooperation of those who control them. Platforms could also manipulate metrics to make them appear more fair than they actually are. Thus, it is important to study methods in which users or communities can intervene in these systems externally to make the AI more aligned with their needs, accountable to those affected, and resilient against harmful optimization dynamics.

Research Interests

5 topics

Algorithmic Collective Action

Privacy-Preserving Systems

Complaint Resolution Systems

AI Ethics & Governance

Natural Language Processing

Selected Publications

peer-reviewed

2026

Conference · AAAI/ACM AIES 2026 PUBLISHED

Test-Time Collective Action: Proxy-Based Perturbations for Correcting Algorithmic Harms

Bhange, M., Aïvodji, U., & Creager, E.

Proc. AAAI/ACM Conference on AI, Ethics, and Society, 9(1), 423–434 · doi:10.1609/aies.v9i1.43152

READ →

2026

Conference · ACM FAccT 2026 PUBLISHED

Crowding Out the Noise: Algorithmic Collective Action Under Differential Privacy

Solanki, R., Bhange, M., Aïvodji, U., & Creager, E.

Proc. 2026 ACM Conference on Fairness, Accountability, and Transparency, pp. 5430–5455 · doi:10.1145/3805689.3806475

READ →

2025

Journal · Computational Intelligence

Survey on AI Ethics: A Socio-Technical Perspective

Mbiazi, D., Bhange, M., Babaei, M., Sheth, I., Kenfack, P., & Kahou, S. E.

Computational Intelligence, 41(6), e70149 · doi:10.1111/coin.70149

READ →

2020

Workshop · SemEval 2020

HinglishNLP at SemEval-2020 Task 9: Fine-tuned Language Models for Hinglish Sentiment Detection

Bhange, M., & Kasliwal, N.

Proc. 14th Workshop on Semantic Evaluation · pp. 934–939

READ →

Pre-prints & Workshop Papers

non-archival

2025

Workshop · ACA @ NeurIPS (non-archival) ORAL

Conscious Data Contribution via Community-Driven Chain-of-Thought Distillation

Libon, L., Bhange, M., Solanki, R., Creager, E., & Aïvodji, U.

Algorithmic Collective Action Workshop, NeurIPS 2025 · arXiv:2512.18174

READ →

meghana.history

$ git log --all

meghana@meghanabhange.com:~/history — zsh — 120×32

⌘T

meghana@meghanabhange.com:~/history$ git log --branch=meghana.history

* 5212bf4 — (OCT '26)

paper: Our paper is published at AAAI/ACM AIES 2026 PUBLISHED

Test-Time Collective Action: Proxy-Based Perturbations for Correcting Algorithmic Harms Meghana Bhange, Ulrich Aïvodji, Elliot Creager doi →

* 4f37cf9 — (JUN '26)

paper: Our paper is published at ACM FAccT 2026 PUBLISHED

Crowding Out the Noise: Algorithmic Collective Action Under Differential Privacy Rushabh Solanki, Meghana Bhange, Ulrich Aïvodji, Elliot Creager doi →

* 9f0d296 — (JUN '26)

log: We're going to be hosting a CRAFT session on collective action & LGBTQIA2S+ community needs at FAccT 2026

Closing the Gap Between Algorithmic Collective Action Research and LGBTQIA2S+ Community Needs — Common Ground, June 25, 2026, Le Centre Sheraton Montréal. event →

* 16d52f4 — (DEC '25)

paper: Our paper got accepted as a poster at ACA @ NeurIPS 2025 POSTER

Crowding Out the Noise: Algorithmic Collective Action Under Differential Privacy Rushabh Solanki, Meghana Bhange, Ulrich Aïvodji, Elliot Creager arXiv →

* a804f72 — (DEC '25)

paper: Our paper got accepted as an oral presentation at ACA @ NeurIPS 2025 ORAL

Conscious Data Contribution via Community-Driven Chain-of-Thought Distillation Lena Libon, Meghana Bhange, Rushabh Solanki, Elliot Creager, Ulrich Aïvodji arXiv →

* 3647bd4 — (DEC '25)

log: Student organizer for ACA Workshop @ NeurIPS 2025

acaworkshop.github.io.

* 4cff62e — (OCT '25)

talk: Gave a talk at IVADO Digital Futures 2025

AI for Working with the Machines: Combating Algorithmic Harms with Collective Action event →

* 57a13dd — (MAY '25)

talk: Presented at 92e Congrès de l'Acfas

Actions collectives algorithmiques : vers une gouvernance participative de l'IA — Colloque 428. event →

* 4e5e5b6 — (MAY '25)

role: PhD Researcher @ ÉTS & Mila

Co-supervised by Prof. Ulrich Aïvodji and Prof. Elliot Creager. Algorithmic Collective Action · Protective Optimisation Technologies.

* 769ab5e — (MAY '25)

log: Starting my PhD at ÉTS Montréal

Co-supervised by Prof. Ulrich Aïvodji and Prof. Elliot Creager, affiliated with TISL Lab and Mila — Québec AI Institute.

* e1e6232 — (APR '25)

log: Completed my Master's at ÉTS Montréal

Finished my M.Eng. in Software and Information Technology Engineering (Winter 2025). Thesis on privacy-preserving complaint resolution using Attribute-Based Credentials and zero-knowledge proofs.

* 97089a0 — (APR '25)

award: Received ÉTS Dean's Academic Scholarship

$1,000 award from the Dean of Studies at École de technologie supérieure for academic performance.

* a19aab6 — (2025)

paper: Journal paper published in Computational Intelligence

Survey on AI Ethics: A Socio-Technical Perspective Dave Mbiazi, Meghana Bhange, Maryam Babaei, Ivaxi Sheth, Patrik Kenfack, Samira Ebrahimi Kahou doi →

* 533b2b8 — (2023)

award: Top essay at Kaggle AI Report 2023

Exploring the Landscape of AI Ethics — selected among global submissions.

* e3a1f63 — (MAY '23)

role: Master's Researcher @ ÉTS

Privacy-preserving complaint resolution using Attribute-Based Credentials and zero-knowledge proofs. Supervised by Prof. Aïvodji & Prof. Robert.

* 33f3d27 — (JAN '23)

role: Freelance ML Engineer @ Toptal

NLP pipelines for inclusive hiring (SpaCy, GPT-3/4, FastAPI) · Algorithmic equity trading platform (Interactive Brokers).

* 11781ae — (2022)

award: Best All-Female Team at UN PET Lab Hackathon

Built a Privacy Enhancing Technologies solution using differential privacy.

* 556ed90 — (2022)

award: 2nd Place at Toptal Hackathon

AI-assisted Cards Against Humanity using Stable Diffusion.

* 4b26c38 — (SEP '21)

role: ML Engineer @ Hunters Security

Analytical ML tools for threat-hunting detectors and anomalous user behaviour across Snowflake and AWS RDS.

* 55a2fb0 — (JUL '20)

role: ML Engineer @ Verloop.io

Intent recognition with sentence-transformers (+40% accuracy) · Multilingual NER service deployed across clients.

* ef7662f — (JUL '19)

talk: Spoke at PyData Bangalore Meetup

Extracting Names from Multi-lingual Conversations — Flair-based multilingual NER. watch →

✓ 21 commits | branch: meghana.history

meghana@meghanabhange.com:~/history$

Let's connect.

Open to collaborations on algorithmic collective action, privacy-preserving ML, and socio-technical systems. Always happy to chat.

EMAILmeghanabhange@hey MASTODON@aspiringcat GITHUBmeghanabhange

References

Hardt, M., Mazumdar, E., Mendler-Dünner, C., & Zrnic, T. (2023). Algorithmic collective action in machine learning. International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA, Proceedings of Machine Learning Research, 202, 12570–12586. https://proceedings.mlr.press/v202/hardt23a.html

Kulynych, B., Overdorf, R., Troncoso, C., & Gürses, S. F. (2020). POTs: Protective optimization technologies. FAT* ’20: Conference on Fairness, Accountability, and Transparency, Barcelona, Spain, January 27-30, 2020, 177–188. https://doi.org/10.1145/3351095.3372853

Shelby, R., Rismani, S., Henne, K., Moon, Aj., Rostamzadeh, N., Nicholas, P., Yilla-Akbari, N., Gallegos, J., Smart, A., Garcı́a, E., & Virk, G. (2023). Sociotechnical harms of algorithmic systems: Scoping a taxonomy for harm reduction. Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society, AIES 2023, Montréal, QC, Canada, August 8-10, 2023, 723–741. https://doi.org/10.1145/3600211.3604673