ResLearn-XR: Residual Learning for Network Traffic and Quality-of-Experience-Aware Modeling in Extended Reality

SummaryAI generated

On September 7, 2026, arXiv proposed ResLearn-XR, a residual learning framework for XR scenarios. This framework aims to predict network traffic and estimate QoE risks. It adopts a two-stage temporal structure, combining a basic sequence prediction model with task-specific residual components to handle the value space and logits space respectively, in order to address the burstiness and non-stationary nature of XR traffic. The study introduced a data description algorithm (DDA) to convert packet-level features in encrypted traffic into frame-aware descriptors, and constructed an XR dataset including continuous traffic trajectories and session-level user reports. Experimental results show that ResLearn-XR reduced SMAPE by up to 17.84% in terms of frame count, frame size, and arrival interval prediction, and reduced SMAPE by up to 87.8% in QoE risk estimation compared to single-stage baselines.

Related eventsRELATED EVENTS

2026-09-09 05:29OpenAI claims huge maths breakthrough on a famed ‘Millennium Problem’
2026-09-09 06:11OpenAI claims to have solved the Navier-Stokes equation problem, but its originality is questioned by the academic community
2026-09-08 22:00Google DeepMind Maps 9 Billion Possible DNA Variants

Key entitiesKEY ENTITIES

ResLearn-XR

SignalsSIGNALS

Keyword heat

ResLearn-XR1

All reports (1)SOURCES

↗

A arXiv cs.AI en 2026-09-07 12:00

arXiv:2609.04493v1 proposes ResLearn-XR, a residual learning framework for predicting XR network traffic and estimating QoE risks. This framework adopts a two-stage temporal learning structure, including a basic sequence prediction model and task-specific residual learning components, operating in the value space and logits space respectively to handle the bursty and non-stationary characteristics of XR traffic. The study introduced a data description algorithm (DDA) that converts observable features at the packet-level application layer into frame-perception descriptors suitable for encrypted traffic analysis, and constructed an XR dataset of paired continuous traffic trajectories and session-level user reported QoE labels. Experiments showed that ResLearn-XR reduced SMAPE by up to 17.84% in terms of frame count, frame size, and arrival interval predictions, and decreased SMAPE for QoE risk estimation by up to 87.8% compared to single-stage baselines.