About Me

I am a Ph.D. candidate in Moon Lab at the Korea Advanced Institute of Science and Technology (KAIST), South Korea, advised by Prof. Jaekyun Moon. My research focuses on making deep learning efficient in data, compute, and communication.

A central thread of my work is data-efficient learning through active learning and data selection. In PruneFuse (TMLR’26 Featured Certification, ICML’26), we select training data efficiently through weight pruning and network fusion. I also work on communication-efficient federated learning, from evolutionary strategies in EvoFed (NeurIPS’23) to model-agnostic projection optimization and consistent low-rank aggregation for federated LoRA fine-tuning. More recently, I have been exploring model merging, including spectral-sparse merging of specialized models (Ada-HiSS).

I am a people person at heart and love building community around research. I co-organize the AIM4DK workshop at IJCAI-ECAI’26, volunteer and review for Women in Machine Learning (WiML) at NeurIPS, and have served as Vice President of the Pakistan Student Association (PSA) at KAIST and as an International Student Mentor. I am always happy to chat with students finding their way into Machine Learning and academia, so feel free to reach out!

Outside the lab, you’ll find me snowboarding, picking up new languages, or swapping life stories over coffee.

📰 News

• Sep 2026 – Our new preprint, “Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning”, is now available on arXiv.
• July 2026 – Our paper, “Router-Guided Data Selection for Efficient Deep Learning” (ICML 2026 workshop), received the Best Student Paper Award!
• July 2026 – Presented “PruneFuse: Efficient Data Selection via Weight Pruning and Network Fusion” at ICML 2026 in Seoul, South Korea.
• May 2026 – Our paper, “PruneFuse: Efficient Data Selection via Weight Pruning and Network Fusion”, received a Featured Certification (Spotlight) in TMLR, accepted to be presented in ICML 2026.
• May 2026 – Two papers accepted to ICML workshop 2026.
• March 2026 – Our Workshop, “AI Modeling for Disappearing Knowledge (AIM4DK)”, has been accepted in IJCAI-ECAI’26.
• Feb 2026 – Successfully passed my Ph.D. Dissertation Proposal Defense!

🏆 Awards & Honors

• Best Student Paper Award, ICML 2026 Workshop, for “Router-Guided Data Selection for Efficient Deep Learning” (2026)
• Featured Certification (Spotlight), Transactions on Machine Learning Research (TMLR), for “PruneFuse: Efficient Data Selection via Weight Pruning and Network Fusion” (2026)

📚 Publications

• NeurIPS’24 WSPruning-based Data Selection and Network Fusion for Efficient Deep Learning
  Humaira Kousar, HI Bhatti, J Moon
  Neural Information Processing Systems (NeurIPS), Workshop on Attributing Model Behavior at Scale, 2024
  Pdf Bibtex

• TMLR’26PruneFuse: Efficient Data Selection via Weight Pruning and Network Fusion
  Humaira Kousar, HI Bhatti, J Moon
  Transactions on Machine Learning Research (TMLR), 2026
  Pdf Bibtex Featured Certification (Spotlight)

• ICML’26 WSRouter-Guided Data Selection for Efficient Deep Learning
  Humaira Kousar, HI Bhatti, MM Rahimi, J Moon
  International Conference on Machine Learning (ICML), Workshop, 2026
  Bibtex Best Student Paper Award

• Under ReviewOvercoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning
  MI Hossain, Humaira Kousar, IC Tourni
  arXiv preprint, 2026 (under review)
  Pdf Bibtex

• Under ReviewConsistent Low-Rank Aggregation for Federated LoRA Fine-Tuning
  MM Rahimi, M Heidari, Humaira Kousar, D Seo, J Moon
  Neural Information Processing Systems (NeurIPS), 2026 (under review)
  Bibtex

• Under ReviewModel-Agnostic Projection Optimization for Communication-Efficient Federated Learning
  MM Rahimi, Humaira Kousar, DJ Han, D Seo, J Moon
  Neural Information Processing Systems (NeurIPS), 2026 (under review)
  Bibtex

• Under ReviewAda-HiSS: Spectral-Sparse Model Merging in the Small-Room Regime
  HI Bhatti, Humaira Kousar, Jaekyun Moon
  Neural Information Processing Systems (NeurIPS), 2026 (under review)
  Bibtex

• NeurIPS’23EvoFed: Leveraging Evolutionary Strategies for Communication-Efficient Federated Learning
  MM Rahimi, HI Bhatti, Y Park, Humaira Kousar, J Moon
  Advances in Neural Information Processing Systems (NeurIPS), 2023
  Pdf Bibtex Code

• ICLR’23 WSDistribution Aware Active Learning via Gaussian Mixtures
  Y Park, DJ Han, J Park, W Choi, Humaira Kousar, J Moon
  International Conference on Learning Representations (ICLR), Workshop on Pitfalls of limited data and computation for Trustworthy ML, 2023
  Pdf Bibtex

• ICGHIT’19WHEGSTR: A Multiterrain Robot with C-Shaped Whegs, Implementation of Error Minimization Technique and using Artificial Neural Network (ANN)
  Humaira Kousar, M Atif, H Ali, MH Kundi
  International Conference on Green and Human Information Technology (ICGHIT), IEEE, 2019
  Bibtex

📚 Patent

Device and Method for Distribution Aware Active Learning using Gaussian Mixture. Application Number: KR 10-2024-0035374, Application Date: Mar 13, 2024

📚 Research Projects

Active Learning Algorithms for Efficient Utilization of Unlabeled Defense Data. Center for Applied Research in Artificial Intelligence (CARAI), Funded by Agency for Defense Development (ADD); (2021 – 2026)

📚 Academic Service

• Organizer: AIM4DK: AI Modeling for Disappearing Knowledge, workshop at IJCAI-ECAI’26
• Conference Reviewer: ICLR’26–27, NeurIPS’26, AAAI’26, EACL’26
• Workshop Reviewer: NeurIPS (WiML’25–26, GlobalSouthAI’26, ATTRIB’24), ICML (AI4GOOD’26, GlobalSouthML’26, AdaptFM’26)
• Volunteer: NeurIPS WiML’25–26