The field of adaptive video streaming and interpretable AI is moving towards personalized optimization and improved comprehensibility. Researchers are exploring new methods to align user-level Quality of Experience with algorithmic optimization objectives, such as using large language models to evaluate decision trees for developer comprehensibility. Additionally, there is a growing interest in sparse autoencoders and their applications in feature extraction and recommendation systems. Noteworthy papers in this area include: Towards User-level QoE, which proposes a large-scale deployed system for personalized adaptive video streaming based on user-level experience, achieving a 0.15% increase in total viewing time and a 1.3% reduction in stall time. Beyond Interpretability, which introduces a bitrate adaptation algorithm generation framework that considers comprehensibility, significantly improving comprehensibility while maintaining competitive performance.
Advances in Adaptive Video Streaming and Interpretable AI
Sources
Towards User-level QoE: Large-scale Practice in Personalized Optimization of Adaptive Video Streaming
Beyond Interpretability: Exploring the Comprehensibility of Adaptive Video Streaming through Large Language Models
Sparse but Wrong: Incorrect L0 Leads to Incorrect Features in Sparse Autoencoders
Attention Layers Add Into Low-Dimensional Residual Subspaces
Opening the Black Box: Interpretable Remedies for Popularity Bias in Recommender Systems
AdaptiveK Sparse Autoencoders: Dynamic Sparsity Allocation for Interpretable LLM Representations
S2Sent: Nested Selectivity Aware Sentence Representation Learning
Sparse Autoencoders for Low-$N$ Protein Function Prediction and Design
Improving Recommendation Fairness via Graph Structure and Representation Augmentation