Breaking the Blind Loop: Cross-Layer Co-Design for Next-Generation Interactive Systems
COM1 Level 3
MR1, COM1-03-19
Abstract:
The migration from TCP to QUIC (HTTP/3) promised lower latency and improved Quality of Experience (QoE), yet in practice, these gains remain limited. The root cause is a fundamental architectural constraint, which I call the "Blind Loop": the application is blind to network conditions, the transport layer is blind to application-level QoE, and the network operates without visibility into either.
In this talk, I will present my Lab’s vision for a symbiotic, cross-layer architecture designed to shatter these silos across the application, transport, edge, and compute layers. The talk begins at the transport layer, demonstrating why QUIC alone is insufficient in highly dynamic network environments. I will show how integrating cross-layer telemetry with learning-based congestion control enables transport protocols to adapt directly to application-level QoE. Scaling this infrastructure to the edge, I will then discuss how we optimize performance and cost-efficiency using multi-agent reinforcement learning for real-time Multi-CDN orchestration. Next, we extend this architecture to immersive six-degrees-of-freedom (6DoF) environments, where dynamic point clouds introduce fundamentally new challenges for adaptive streaming. To solve this, I will present a complete end-to-end volumetric ecosystem featuring a unified content preparation toolkit, a high-speed simulation platform, a browser-native volumetric media player, and a large-scale dataset with user navigation traces, enabling fast, reproducible, and behavior-aware optimization of volumetric streaming systems. We will then address the compute bottlenecks of AI-driven video enhancement through VSR-Bench, the first browser-native platform for evaluating video super-resolution within live WebRTC pipelines. Our findings expose critical trade-offs between perceptual quality, latency, and resource utilization. Finally, I will introduce a new paradigm that tightly integrates networking with AI inference for Multimodal LLMs. By pipelining progressive video transport directly into incremental LLM processing, we break the receive-then-process barrier, eliminating GPU cold-start overheads and drastically reducing interactive end-to-end latency.
Ultimately, this talk demonstrates that the future of ultra-low-latency interactive systems cannot be achieved through isolated optimizations. It requires a holistic co-design where networking, edge infrastructure, and AI compute operate as a unified, seamlessly integrated stack.
Bio:
Dr. Abdelhak Bentaleb is a Gina Cody Research and Innovation Fellow and Assistant Professor in the Department of Computer Science and Software Engineering at Concordia University, Montreal, Canada. He is the Founder and Director of the IN2GM Lab, where he leads research on next-generation intelligent multimedia and networking systems. He received his Ph.D. in Computer Science from the National University of Singapore in 2019, where he subsequently served as a Research Fellow until 2022.
Dr. Bentaleb’s research lies at the intersection of networking, multimedia, systems, and applied artificial intelligence. His work focuses on designing data-driven, scalable, and energy-efficient systems for video streaming, immersive media (e.g., 6DoF and volumetric video), content delivery, and next-generation Internet architectures. He is particularly interested in AI-driven optimization of end-to-end systems, transport protocols such as QUIC, edge/cloud intelligence, and sustainable networking for beyond-5G/6G environments. His research is supported by major industry partners, and he is currently leading and collaborating on several R&D projects with organizations such as InterDigital, Ericsson, and TELUS, among others. He is the recipient of several prestigious awards, including the SIGMM Award for Outstanding PhD Thesis, the DASH-IF Best PhD Dissertation Award, and the NUS Dean’s Graduate Research Excellence Award (AY2018/2019). More information can be found at: https://ddec1-0-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2fusers.encs.concordia.ca%2f%7eabentale%2f&umid=16a3836a-78ec-4845-8698-e107b0fc96ef&rct=1774748071&auth=8d3ccd473d52f326e51c0f75cb32c9541898e5d5-4ec21695e58049f32e35d4aee9745ed74cd70a03