Weekly Overview: This Week in Computer Science Papers – December 15, 2025

Key Takeaway

The weekly overview “This Week In Computer Science Papers” showcases a broad range of current research papers, with a strong focus on Computer Vision, Artificial Intelligence, and Hardware Verification.

Summary

• Papers Overview
– Total papers displayed: 36 of 345 that week.
– Main categories and counts:

• Machine Learning (cs.LG) – 108
• Computer Vision and Pattern Recognition – 106
• Artificial Intelligence – 77
• Computation and Language – 43
• Robotics (cs.RO) – 33
• Cryptography and Security – 19
• Information Retrieval – 17
• Computers and Society – 12
• Information Theory (cs.IT) – 12
• Software Engineering – 12
• Distributed, Parallel, and Cluster Computing – 11
• Hardware Architecture – 8
• Graphics (cs.GR) – 7
• Human-Computer Interaction – 7
• Computational Complexity – 6
• Computational Engineering, Finance, and Science – 5
• Data Structures and Algorithms – 5
• Computer Science and Game Theory – 4
• Multiagent Systems (cs.MA) – 4
• Sound (cs.SD) – 4
• Computational Geometry – 3
• Discrete Mathematics (cs.DM) – 3
• Logic in Computer Science – 3
• Networking and Internet Architecture – 3
• Databases (cs.DB) – 2
• Emerging Technologies – 2
• Multimedia (cs.MM) – 2
• Performance (cs.PF) – 2
• Social and Information Networks – 2
• Digital Libraries (cs.DL) – 1
• Formal Languages and Automata Theory – 1
• Mathematical Software – 1
• Neural and Evolutionary Computing – 1
• Operating Systems (cs.OS) – 1
• Programming Languages – 1
• Symbolic Computation – 1

Highlighted Papers (since December 15, 2025)

1. DiffusionBrowser – Interactive preview of video diffusion models: lightweight decoder, >4× real‑time, allows control of generation process via stochastic injection and modal control. Systematic analysis of the denoising process.
2. LitePT – Lightweight, more powerful Point‑Transformer: convolution in early layers, attention in deep layers, combined with PointROPE‑Encoding. 3.6× fewer parameters, 2× faster, 2× less memory, matches or surpasses Point‑Transformer‑V3.
3. VTP (Visual Tokenizer Pre‑training) – Scalable pre‑training for visual tokenizers: combines image‑text contrastive, self‑supervised, and reconstruction losses. Produces semantically compact latent spaces, shows significant gains in generation (FID optimisation, faster convergence).
4. Lyra – Hardware‑accelerated RISC‑V verification framework: FPGA‑SoC with ISA‑aware generative model (LyraGen). Increases coverage by 27 % and speeds up verification 107–3343× vs software fuzzer.
5. Semantic Analysis for Alzheimer’s Diagnosis – Pipeline to examine semantic cues in speech samples: syntax‑ and vocabulary‑transformed texts retain semantics, models can detect AD even with major structural changes.
6. Recurrent Video Masked Autoencoders (RVM) – Recurrent transformer for spatio‑temporal video representations: efficient masked prediction, 30× parameter‑efficient compared to VideoMAE/V‑JEPA, stable feature propagation over long horizons.

How the Overview Works

– Hover tile shows abstract preview; click opens detail page.
– Sidebar filters enable search by category.
– Highlights progress in ML, Vision, AI and Hardware.

Related Queries

Welche Fortschritte wurden mit DiffusionBrowser bei der Interaktion mit Video‑Diffusion-Modellen gemacht?
Wie verbessert LitePT die Effizienz und Performance von 3D‑Point‑Cloud‑Netzwerken?
Welche Erkenntnisse liefert VTP hinsichtlich des Pretrainings von Visual Tokenizers für generative Modelle?

Source: https://www.weekinpapers.com/