RESEARCH

Our research spans inter-related disciplines of computer science and engineering with a focus on system design and optimisation for machine learning systems (training and inference efficiency, distributed ML, federated learning), distributed systems (architecture design, performance analysis, resource allocation, algorithmic optimisation), computer networks (traffic engineering, congestion control, performance optimisation, software-defined networking), and wireless networks. I lead the UKRI-EPSRC project KUber to reshape knowledge-delivery systems at scale with the goal of democratising AI.

Current Research Interests

• Systems for ML & ML for Systems

• Edge AI, Federated/Distributed ML and Privacy-Preserving AI

• Efficient and Resource-Constrained Generative AI, LLMs and Agentic AI

• Internet of Things, Digital Twins and Sustainable Computing

• Computer Networks and Distributed Systems

Other Research Interests

• Congestion Control, Resource Allocation and Traffic Engineering

• Software Defined Networking and Network Function Virtualization

• Feedback Control of Hybrid and Switched Systems

• Mobile Ad-hoc Networks and Wireless Sensor Networks

• AI Optimization and Genetic Algorithms

Optimizing Distributed Machine Learning Applications

We are exploring and working on subjects that fall on the intersection between Computer/Network Systems Research and Machine Learning, including resource-efficient federated learning, communication compression for distributed training, knowledge transfer across heterogeneous learners, and efficient fine-tuning and serving of large models. See the Projects and Publications pages for details.

Cloud Computing and Data Center Networks

We are working extensively on problems related to applications’ performance issues in the non-conventional data center networking environments which differs completely from issues applications’ face in the wild internet. Our primary focus is public could and hence we develop techniques which imposes no modifications to the guest VMs in public clouds. The directions i am focusing under this broad category are exploring the following venues:

• Switch-based approaches to resolve the problems from within the network itself.

• Hypervisor-based approaches to resolve the problems from within the hypervisor of the host OS.

• SDN-based approaches to resolve the problems from within the controller of the network.

Switch based Approaches

In switch-based approaches we aim for developing intelligent Active Queue Management schemes that are designed specifically for data center environments. The schemes shall be simple enough for real adoption and incremental deployment in production large-scale data centers.

1. RWNDQ: Receive-window based Active Queue Management.

2. IQM: Incast-aware Active Queue Management.

3. HSCC: Smart Hysteresis-based Active Queue Management.

Hypervisor-based Approaches

Full list of publications.

SDN-based Approaches

Full list of publications.