Electrical and Computer Engineering · Concentration in Artificial Intelligence ·GPA 4.0/4.0
Atlanta, GA
Thesis Explicit Group Sparse Projection for Machine Learning
Experience
Google DeepMind Research Scientist
Jul 2026 – Present
Permanent, full-time
New York City, NY
Research on representations, control, alignment and interpretability of large-scale frontier models.
Google DeepMind Research Intern
Sep 2025 – May 2026
Model Alignment, Control, Responsibility and Safety
Kirkland, WA
Studied how concepts are represented in the residual stream of diffusion transformers, and how that structure can be used to steer generation (CVPR HOW workshop 2026).
Cohere Intern of the Technical Staff
Sep 2024 – Dec 2024
LLM Efficiency Research
Atlanta, GA
Developed inference-time activation sparsity methods for large language models, targeting deployment settings where latency and serving cost dominate.
Dolby Laboratories PhD Research Intern
May 2024 – Aug 2024
Experience Delivery Lab, Advanced Technologies Group
Atlanta, GA
Designed a parameter-efficient fine-tuning method for LLMs based on probabilistic layer selection, concentrating updates on the layers most relevant to the downstream task.
Meta AI (FAIR) Research Scientist Intern
May 2022 – Aug 2022
Sparsity and Efficiency
Menlo Park, CA
Built WeiGit, a git-like library for versioning and compressing neural network weights during training; released as part of facebookresearch/fairscale.
Studied extreme sparsity in deep networks using signal-processing transforms (FFT, DCT) applied during training.
TReNDS Center Graduate Research Assistant
Aug 2019 – Aug 2026
Georgia Institute of Technology
Atlanta, GA
Introduced sparse adapters as parameter-efficient building blocks for modular architectures, outperforming LoRA and full fine-tuning when merged across 20 NLP tasks; with Mila, Montreal (COLM 2025).
Developed communication-efficient federated learning for the non-IID regime, applied to distributed neuroimaging (TMLR 2026; Frontiers in Neuroinformatics 2024).
Designed single-shot pruning and neural-pathway discovery for offline and multi-task reinforcement learning; with Mila (NeurIPS 2024).
Built Grouped Sparse Projection, a projection-based algorithm that trains sparse deep models to near-baseline accuracy above 90% sparsity (TMLR 2022; ICLR HAET workshop 2021).
Publications
R. Ohib, M. Hahn, M. Malek. Concept Spaces in the Residual Stream of Diffusion Transformers.IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), HOW workshop, 2026. [paper]
J. Gammell, B. Thapaliya, Y. Jung, R. Ohib, B. Fehri, D. Chakrabarti. Learning to Query History: Nonstationary Classification via Learned Retrieval.International Conference on Learning Representations (ICLR), TSALM workshop, 2026. [paper]
R. Ohib, B. Thapaliya, G. K. Dziugaite, J. Liu, V. D. Calhoun, S. Plis. SSFL: Discovering Sparse Unified Subnetworks at Initialization for Efficient Federated Learning.Transactions on Machine Learning Research (TMLR), 2026. [paper]
S. Y. Arnob, Z. Su, M. Kim, O. Ostapenko, R. Ohib, E. Saleh, D. Precup, L. Caccia, A. Sordoni. Exploring Sparse Adapters for Scalable Merging of Parameter Efficient Experts.Conference on Language Modeling (COLM), 2025. [paper]
B. Thapaliya*, R. Ohib*, E. Geenjaar, J. Liu, V. Calhoun, S. M. Plis. Efficient Federated Learning for Distributed Neuroimaging Data.Frontiers in Neuroinformatics, 2024. [paper]
S. Y. Arnob, R. Ohib, S. M. Plis, A. Zhang, A. Sordoni, D. Precup. Efficient Reinforcement Learning by Discovering Neural Pathways.Conference on Neural Information Processing Systems (NeurIPS), 2024. [paper]
B. Thapaliya, R. Ohib, E. Geenjaar, J. Liu, V. Calhoun, S. M. Plis. Decentralized Sparse Federated Learning for Efficient Training on Distributed Neuroimaging Data.Conference on Neural Information Processing Systems (NeurIPS), MedNeurIPS workshop, 2023. [paper]
R. Ohib, B. Thapaliya, P. Gaggenapalli, J. Liu, V. Calhoun, S. Plis. SalientGrads: Sparse Models for Communication Efficient and Data Aware Distributed Federated Training.International Conference on Learning Representations (ICLR), SNN workshop, 2023. [paper]
E. Geenjaar, D. Kim, R. Ohib, M. Duda, A. Kashyap, S. M. Plis, V. Calhoun. Uncovering the latent dynamics of whole-brain fMRI tasks with a sequential variational autoencoder.Conference on Neural Information Processing Systems (NeurIPS), DeepGen workshop, 2023. [paper]
R. Ohib, N. Gillis, N. Dalmasso, S. Shah, V. Potluru, S. Plis. Explicit Group Sparse Projection with Applications to Deep Learning and NMF.Transactions on Machine Learning Research (TMLR), 2022. [paper]
R. Ohib, N. Gillis, S. Shah, V. Potluru, S. Plis. Grouped Sparse Projection for Deep Learning.International Conference on Learning Representations (ICLR), HAET workshop, 2021. [paper]
S. Yeasar, R. Ohib, S. Plis, D. Precup. Single-Shot Pruning for Offline Reinforcement Learning.Conference on Neural Information Processing Systems (NeurIPS), Off-RL workshop, 2021. [paper]
R. Ohib, S. Y. Arnob, M. Muhaisin, R. Arefin, T. Reza, M. R. Amin. ENF Based Machine Learning Classification for Origin of Media Signals: Novel Features from Fourier Transform Profile.International Conference on Electrical, Electronics and Computer Science (ICEECS), 2018.
S. Y. Arnob, R. Ohib, M. M. Muhaisin, T. B. Hassan. Power file extraction process from bangladesh grid and exploring ENF based classification accuracy using machine learning.2017 IEEE Region 10 Humanitarian Technology Conference (R10-HTC), 2017. [paper]
* Equal contribution
Talks and presentations
The Hidden Structure of Deep Models: Discovering and Exploiting Sparse Subnetworks, TReNDS Center, Atlanta, GA
Jul 2026
What Models Learn and What They Hide: Sparse Structures and Representation Control, Google DeepMind, Kirkland, WA
Apr 2026
Navigating a PhD and Doing Scientific Research, TReNDS Center, Atlanta, GA
Apr 2026
Open source
WeiGitWeight versioning for neural networks·Meta AI (FAIR)·facebookresearch/fairscaleVersion control and compression for model weights during training, exposed through a git-like command line and Python API; merged into the fairscale library.
2022
Honors and awards
Full Funding Award, CIFAR DLRL Summer SchoolSelected by research proposal·CIFAR and Amii·Edmonton, Canada
2019
Honorable Mention, IEEE Signal Processing CupTeam and programming lead
2016
Undergraduate Honors ListNamed to the Graduating Honors Group for academic excellence
2013 – 2017
OIC ScholarshipFull tuition waiver and monthly stipend for the duration of undergraduate study
2013 – 2017
Academic service
Program CommitteeInternational Semantic Web Conference (ISWC)
2026
ReviewerConference on Neural Information Processing Systems (NeurIPS)International Conference on Learning Representations (ICLR)Transactions on Machine Learning Research (TMLR)IEEE Transactions on Neural Networks and Learning Systems (TNNLS)
2023 – Present
Graduate coursework
Statistical Machine Learning
Convex Optimization
Linear Algebra
Advanced Digital Signal Processing
Advanced Programming Techniques
Fourier Analysis
Information Processing in Neural Systems
Real Analysis
Programming
Languages Python, C++, Matlab, Bash
Frameworks PyTorch, JAX, NumPy, Pandas
Infrastructure Linux, Slurm, distributed and multi-GPU training, cluster computing
Publications, research interests and contact details are in the
full CV (PDF).