CNI CONNECT

Monthly Newsletter from CNI

Issue - February 2026

Upcoming Talks

As the semester gathers pace, the CNI seminar series continues with another strong lineup of speakers through February and March. With voices from across academia and industry, in India and beyond, the coming weeks promise a wide range of ideas and conversations.

Here’s what’s ahead:

Feb 25: Alexandre Reiffers-Masson, IMT Atlantique
Mar 02: Hariprasad Manjunath Hegde, Chanakya University
Mar 04: Sandeep Chennakeshu, Unhunder

Mar 09: Bharadwaj Satchidanandan, IIT Madras
Mar 16: Ashutosh Sabharwal, Rice University
Mar 23: Rijurekha Sen, IIT Delhi

Mar 30: Dheeraj Nagaraj, Google

Talk titles, abstracts, and full details are available on the CNI Seminar Series page. Do take a look and mark your calendars.

CNI is inviting applications for the following openings in the Distributed Systems Lab:

1. Research Associate: For experienced professionals and fresh graduates planning to get into systems research.

2. Research Intern: For students who are interested in experimental research in systems/networks while pursuing their studies.

More details at the CNI opportunities page.

Rearchitecting Datacenter Networks: A New Paradigm with Optical Core and Optical Edge

Dr. Sushovan Das (ETH Zürich) presented OSSV, a novel datacenter architecture combining optical circuit-switched core and edge designs. By jointly minimizing traffic skewness and inter-rack volume, the framework improves performance of all-optical DCNs under realistic workloads while maintaining energy efficiency and scalability in post-Moore networking environments.

On Some Ultra-Sharp Bounds in Queueing Systems

Prof. Florin Ciucu (University of Warwick) introduced two new classes of ultra-sharp bounds in queueing theory. He presented exact delay distribution expressions for correlated G/G/1 queues and novel bounds for tandem networks, significantly improving existing results and revealing a polynomial-exponential structure in end-to-end delays.

Learning with Minimal Human Feedback

Prof. Alexandre Proutiere (KTH Royal Institute of Technology, Sweden) examined how to reduce reliance on costly human feedback in online classification tasks. He introduced an expert-guided framework with algorithms for varying feedback budgets, establishing regret guarantees and validating performance through experiments on real-world question-answering datasets.

The Synergy Between Quality of Experience and Deep Reinforcement Learning for Uncoordinated Multi-Agent Resource Allocation in Cognitive Radio Network

Prof. Andres Kwasinski (Rochester Institute of Technology, USA) explored how Quality of Experience metrics can enable transferable learning in distributed cognitive radio networks. By using a unified QoE scale across traffic types, radios can share reinforcement learning insights, accelerating adaptation with minimal performance loss.

Aging Brain Research: Challenges and Opportunities

Prof. K. V. S. Hari (Centre for Brain Research, IISc) discussed mechanisms underlying healthy and diseased aging, highlighting challenges in assessing cognitive decline and dementia. He emphasized multimodal data integration and the potential of genomics, proteomics, and digital biomarkers to enable early detection and timely intervention.

Oneiros: KV Cache Optimization through Parameter Remapping for Multi-tenant LLM Serving

Dr. Ruihao Li (Meta, AI and Systems Co-Design Group) introduced Oneiros, a system that improves LLM inference by remapping model parameter memory to expand KV cache capacity. By avoiding costly CPU–GPU cache swapping, it significantly reduces latency and improves throughput in multi-tenant serving environments.

Rethinking Networking through the Lens of AI

Dr. Puneet Sharma (HPE Labs) reflected on lessons from applying AI to networking through projects such as Maestro and Wixor. Blending technical depth with a few James Bond analogies, he examined problem formulation, model–system tradeoffs, and how AI can reshape the design of intelligent, trustworthy networks.

Reinforcement Learning in Non-Stationary Environments

Prof. Pranay Sharma (CMInDS, IIT Bombay) addressed reinforcement learning in time-varying settings, modeling the problem as an MDP with changing rewards and transitions. He introduced NS-NAC, a policy-based method with restart-driven exploration, and a parameter-free variant, advancing theoretical understanding of non-stationary RL.

AI/ML based 6G Wireless Network Design

Prof. Rohit Budhiraja (IIT Kanpur) presented the architecture of India’s indigenous 5G network, developed in-house and now deployed at multiple sites. He then discussed how AI/ML techniques are shaping 6G design, outlining efforts toward building an AI-native wireless network for the next decade.

Resource-efficient NextG MIMO Sensing and Communications: Bridging Sparse Arrays, Spatial Modulation and Subspace Codes

Prof. Robin Rajamäki (Tampere University, Finland) examined how sparse antenna geometries unify sensing, modulation, and coding. Linking direction-of-arrival estimation with subspace coding, he highlighted structured codes and antenna selection strategies that enhance resolution, energy efficiency, and hardware simplicity in next-generation MIMO systems.

Mathematical ML and ML for Math: Alternating GD and Minimization (AltGDmin) for Secure Federated Low Rank Matrix Learning for Real-time MRI and ML-enabled K-12 Math Support

Prof. Namrata Vaswani (Iowa State University) presented AltGDmin and its Byzantine-resilient variant for secure, communication-efficient federated low-rank learning, with applications including dynamic MRI. She also discussed ML-enabled tools and the CyMath program aimed at strengthening foundational K–12 mathematics learning.