Learning at extreme scale
Low-precision and sparse training for models with millions of labels, supported by efficient GPU kernels. I have also studied probability calibration in extreme multi-label classifiers.
MACHINE LEARNING · SYSTEMS · EFFICIENCY
Making large-scale
learning efficient.
I develop methods that make machine learning models more efficient in memory and compute, with a focus on large output spaces.
I am a doctoral researcher in Computer Science at Aalto University, supervised by Rohit Babbar and Pekka Marttinen. My research focuses on extreme multi-label classification, low-precision training, dynamic sparsity, and efficient language models.
Recently, I was a Deep Learning Research Intern in the Model Efficiency team at Qualcomm AI Research in Amsterdam, supervised by Mart van Baalen and Markus Nagel. I worked on residual quantization, dynamic mixed precision, and efficient caching for mixture-of-experts models on edge devices.
Explore my publicationsResearch directions
01Efficient language models
Model compression, mixed-precision quantization, and speculative decoding for practical inference.
Publications
02EMNLP 2026
HSEARCH-R1: Learning Compact Value Models for LLM-guided Hierarchical Search via RL
ICML 2026
Preprint
DynaSpec: Context-aware Dynamic Speculative Sampling for Large-Vocabulary Language Models
ICML 2025
KDD 2025
Labels in Extremes: How Well Calibrated are Extreme Multi-label Classifiers?
COLING 2025
Large Language Model as a Teacher for Zero-shot Tagging at Extreme Scales
NeurIPS 2024
Navigating Extremes: Dynamic Sparse Training for Large Output Spaces
CVPR Workshops 2022
A Joint Cross-Attention Model for Audio-Visual Fusion in Dimensional Emotion Recognition
ACCV 2022
Thinking Hallucination for Video Captioning
Preprint
Boosting Video Captioning with Dynamic Loss Network
Research & industry
032023 — Present
Doctoral Researcher · Aalto University
Hardware-efficient deep learning for extreme-scale classification, supervised by Rohit Babbar and Pekka Marttinen. Research on low-precision training, sparsity, and large-vocabulary inference.
May — Sep 2026
Deep Learning Research Intern · Qualcomm AI Research
Model Efficiency team, supervised by Mart van Baalen and Markus Nagel. Residual quantization and dynamic mixed-precision compression of mixture-of-experts models, with efficient caching for edge devices. Amsterdam, Netherlands.
Dec 2021 — Dec 2022
Research collaboration · ÉTS Montréal
Audio-visual attention models for emotion recognition in eHealth applications.
Oct 2019 — Dec 2021
Project Linked Person · Indian Statistical Institute
Video understanding and captioning, including hallucination mitigation and dynamic loss functions. ECSU, Kolkata.
Jun 2018 — Jul 2019
Deputy Data Science Manager · Flytxt
Natural language processing for résumé screening and generative modeling of customer data.
Education
042023 — Present
PhD in Computer Science
Aalto University, Finland · Expected 2027
2018
MTech in Machine Learning and Computing
Indian Institute of Space Science and Technology
Thesis: Video to Text in the Wild
Teaching & community
05Teaching & mentoring
- Aalto University
Teaching Assistant, Machine Learning (2023, 2024) and Programming Parallel Computers (2025). - UpGrad / LJMU · 2019–2022
Supervised 50+ master’s theses in Data Science and Machine Learning. - Winter School on Deep Learning · 2022
Tutorials and practical sessions.
Recognition & service
- ICML 2026 · Gold Reviewer
- KDD 2025 · Student Travel Award and Student Volunteer
- Aalto SCI 2023 · Teaching Assistants of the Year
- Reviewer
NeurIPS 2025, ICLR 2026, ICML 2026, ICLR 2027, AAAI 2027, and TMLR.
GET IN TOUCH
Let’s talk research.
For research discussions and collaboration, reach me at
nasibullah.nasibullah@aalto.fi