Channel-Remix
Efficient Task-Specific Knowledge Transfer via Pretrained Channel Remixing
Build compact task-specific models by selecting and sparsely recombining pretrained channels instead of relearning new weight directions.
MBZUAI
Research &
publications.
My work on foundation models, compression, efficient deep learning, adaptation, and meta-learning, as well as AutoML and neural architecture search. Papers are listed newest first.
Filter publications
Filter by topic, conference, organization, or publication status.
CarveVFM
Carving Specialists Out of Generalists: Automatic Task-Specific Compression of Vision Foundation Models
Jointly search task-specific capacity allocation, removal operators, and recoverability to extract a compact dense model.
MBZUAI
MetaVFM
Meta-Learned Efficient Adaptation of Vision Foundation Models for Medical Imaging
Use meta-learning to select and initialize adapters for a new medical imaging task.
MBZUAI
MedNNS
Supernet-based Medical Task-Adaptive Neural Network Search
Embed medical datasets and supernet subnetworks in a shared space, so a new task can retrieve a matching architecture and initialization.
MBZUAI · BioMEDIA Lab
MARVIN
Monitoring Anomalies via Railway Video Insight and Neural Networks
Railway anomaly detection combining self-supervised representations with synthetic defect rendering and deployment on edge hardware.
LAMIH · ALSTOM
DFuseNAS
A Novel Approach to Neural Architecture Search Based on Diffusion Processes
Use diffusion processes to guide continuous neural architecture search and discover efficient model architectures.
LAMIH · IBM Research
HyT-NAS
Hybrid Transformers Neural Architecture Search for Edge Devices
Search hybrid CNN–Transformer architectures under device constraints, connecting model design with low-power edge deployment.
LAMIH · ESI Algiers
Master’s thesis
Neural Architecture Search of Efficient Vision Transformers for Object Recognition on Tiny Edge Devices
Study neural architecture search for efficient Vision Transformers and object recognition on resource-constrained edge devices.
LAMIH · ESI Algiers
MoENAS
Mixture-of-Experts NAS for Accurate, Fair, and Robust Edge DNNs
Combine hardware-aware architecture search and sparse mixture-of-experts routing, considering accuracy, fairness, and robustness for edge deployment.
NYUAD · CQTS
CV-SPINNS
Complex-Valued Spin Physics-Informed Neural Networks for Quantum Simulation
Use complex-valued physics-informed neural networks for quantum spin simulation, embedding physical constraints in the network architecture.
NYUAD · CQTS
MaskVFM
Prompt-Conditioned Structured Masking for Vision Encoders
Ongoing work on a hypernetwork that generates attention and FFN masks for SAM-style vision encoders from prompts and budget constraints.
MBZUAI
No matching work
Try another combination, or reset the filters.
Audio & supporting materials
Project-specific NotebookLM audio links are included where available. These open in NotebookLM and may require sign-in.
Audio collection linked from my original site ↗CONTACT
Contact
You can email me about my research or a possible collaboration.