Highlights of Recent AI Research Papers from Top Conferences (2023)

This document presents a curated list of recent academic papers selected from CCF‑A conferences (KDD2023, WWW2023, CVPR2023, IEEE VR2023, ICLR2023, ACM MM2023, INFORMS2023, SIGIR2023, etc.). The papers cover a broad range of AI‑related topics such as video quality assessment, e‑commerce recommendation, reinforcement learning for vehicle routing, diffusion models for dance generation, multi‑objective re‑ranking, and rate limiting for micro‑services.

CVPR2023 – Video Quality Assessment Based on Swin Transformer with Spatio‑Temporal Feature Fusion and Data Augmentation A Swin‑Transformer‑based model that fuses spatio‑temporal features and applies data augmentation to achieve state‑of‑the‑art performance on two VQA benchmarks and wins the CVPR NTIRE 2023 video‑enhancement challenge.

CVPR2023 – DATE: Domain Adaptive Product Seeker for E‑commerce A dual‑branch cross‑modal architecture that jointly processes visual and audio‑text cues for product localization in live‑stream e‑commerce, combined with unsupervised domain adaptation to reduce annotation cost.

CVPR2023 – MD‑VQA: Multi‑Dimensional Quality Assessment for UGC Live Videos A no‑reference VQA model that extracts semantic, distortion, and motion features from user‑generated live streams and achieves SOTA results on a newly built UGC dataset.

ICLR2023 – Generalize Learned Heuristics to Solve Large‑scale Vehicle Routing Problems in Real‑time A two‑stage transformer‑based splitting method (TAM) that generates sub‑paths instead of node sequences, enabling models trained on small VRP instances to generalize to real‑time solving of problems with thousands of nodes.

ACM MM2023 – DiffDance: Cascaded Human Motion Diffusion Model for Dance Generation A cascaded diffusion framework that first generates coarse motion from music and then refines it with a super‑resolution diffusion model, achieving high‑fidelity, rhythm‑aligned dance sequences.

INFORMS2023 – AI vs. Human Buyers: A Study of Alibaba’s Inventory Replenishment System A deep reinforcement‑learning system combined with fictitious play that outperforms human buy‑side decisions in reducing stock‑outs and inventory levels, especially during pandemic‑induced demand shocks.

Each entry includes a download link to the full PDF for further reading.