@arXiv_csLG_bot@mastoxiv.page2026-02-25 16:08:29
Replaced article(s) found for cs.LG. https://arxiv.org/list/cs.LG/new
[6/6]:
- Fast-ThinkAct: Efficient Vision-Language-Action Reasoning via Verbalizable Latent Planning
Chi-Pin Huang, Yunze Man, Zhiding Yu, Min-Hung Chen, Jan Kautz, Yu-Chiang Frank Wang, Fu-En Yang
https://arxiv.org/abs/2601.09708 https://mastoxiv.page/@arXiv_csCV_bot/115898618760721320
- Universality of Many-body Projected Ensemble for Learning Quantum Data Distribution
Quoc Hoan Tran, Koki Chinzei, Yasuhiro Endo, Hirotaka Oshima
https://arxiv.org/abs/2601.18637 https://mastoxiv.page/@arXiv_quantph_bot/115967001797773134
- FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning
Haozheng Luo, Zhuolin Jiang, Md Zahid Hasan, Yan Chen, Soumalya Sarkar
https://arxiv.org/abs/2601.19001 https://mastoxiv.page/@arXiv_csCL_bot/115972068838908815
- Analysis of Shuffling Beyond Pure Local Differential Privacy
Shun Takagi, Seng Pei Liew
https://arxiv.org/abs/2601.19154 https://mastoxiv.page/@arXiv_csDS_bot/115971701218309765
- CryoLVM: Self-supervised Learning from Cryo-EM Density Maps with Large Vision Models
Weining Fu, Kai Shu, Kui Xu, Qiangfeng Cliff Zhang
https://arxiv.org/abs/2602.02620
- XtraLight-MedMamba for Classification of Neoplastic Tubular Adenomas
Sultana, Afsar, Rahu, Singh, Shula, Combs, Forchetti, Asari
https://arxiv.org/abs/2602.04819
- Flow-Based Conformal Predictive Distributions
Trevor Harris
https://arxiv.org/abs/2602.07633 https://mastoxiv.page/@arXiv_statML_bot/116045671088130364
- GOT-Edit: Geometry-Aware Generic Object Tracking via Online Model Editing
Shih-Fang Chen, Jun-Cheng Chen, I-Hong Jhuo, Yen-Yu Lin
https://arxiv.org/abs/2602.08550 https://mastoxiv.page/@arXiv_csCV_bot/116046486984991360
- UI-Venus-1.5 Technical Report
Venus Team, et al.
https://arxiv.org/abs/2602.09082 https://mastoxiv.page/@arXiv_csCV_bot/116050980295461008
- The Wisdom of Many Queries: Complexity-Diversity Principle for Dense Retriever Training
Xincan Feng, Noriki Nishida, Yusuke Sakai, Yuji Matsumoto
https://arxiv.org/abs/2602.09448 https://mastoxiv.page/@arXiv_csIR_bot/116051022881293649
- Intent Laundering: AI Safety Datasets Are Not What They Seem
Shahriar Golchin, Marc Wetter
https://arxiv.org/abs/2602.16729 https://mastoxiv.page/@arXiv_csCR_bot/116101884238965526
- The Metaphysics We Train: A Heideggerian Reading of Machine Learning
Heman Shakeri
https://arxiv.org/abs/2602.19028 https://mastoxiv.page/@arXiv_csCY_bot/116125225694943789
- Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks
David Schmotz, Luca Beurer-Kellner, Sahar Abdelnabi, Maksym Andriushchenko
https://arxiv.org/abs/2602.20156 https://mastoxiv.page/@arXiv_csCR_bot/116125330557447048
- A Very Big Video Reasoning Suite
Maijunxian Wang, et al.
https://arxiv.org/abs/2602.20159 https://mastoxiv.page/@arXiv_csCV_bot/116125664801070747
toXiv_bot_toot
@cdarwin@c.im2026-04-25 12:40:38
This summer is shaping up to be an expensive and uncertain time to fly, especially if you’re planning a European vacation.
Volatile fuel prices because of the war in Iran are straining airlines around the globe,
but perhaps nowhere as much as in Europe,
where jet fuel supplies could run low by mid-May.
European airlines like Lufthansa and KLM have announced they are cutting flights,
and others could follow.
Carriers everywhere are increasingly passing cost…
@thomasfuchs@hachyderm.io2026-03-24 12:22:12
RE: https://toot.cat/@plexus/116283016837715719
It should also be noted that beyond the ethical, political and environmental issues with this is that it doesn't work:
1. There is on average no mid to long term productivity gain with actual real-world software development that isn't just a "wow see what it can do" demo. (Multiple studies have shown that now.)
2. It won't help with 90% of the work when professionally making software, which, believe it or not, isn't coding. 90% of the work is designing and planning the software (these are things that happen both upfront and during development).
Maybe you have seen the recent Microsoft thing rolling back features in Windows they added?
E.g. Copilot in Notepad. What they did is essentially outsourcing project management to developers who then outsourced it to LLMs. But an LLMs can't plan and design software, and arguably barely can even generate code that works (as in reliable and performant). So now they have a buggy mess with features no one wants and they're rolling it back.
There's no silver bullets in software development.
@arXiv_csPF_bot@mastoxiv.page2026-03-24 07:38:32
Democratizing AI: A Comparative Study in Deep Learning Efficiency and Future Trends in Computational Processing
Lisan Al Amin, Md Ismail Hossain, Rupak Kumar Das, Mahbubul Islam, Saddam Mukta, Abdulaziz Tabbakh
https://arxiv.org/abs/2603.20920 https://arxiv.org/pdf/2603.20920 https://arxiv.org/html/2603.20920
arXiv:2603.20920v1 Announce Type: new
Abstract: The exponential growth in data has intensified the demand for computational power to train large-scale deep learning models. However, the rapid growth in model size and complexity raises concerns about equal and fair access to computational resources, particularly under increasing energy and infrastructure constraints. GPUs have emerged as essential for accelerating such workloads. This study benchmarks four deep learning models (Conv6, VGG16, ResNet18, CycleGAN) using TensorFlow and PyTorch on Intel Xeon CPUs and NVIDIA Tesla T4 GPUs. Our experiments demonstrate that, on average, GPU training achieves speedups ranging from 11x to 246x depending on model complexity, with lightweight models (Conv6) showing the highest acceleration (246x), mid-sized models (VGG16, ResNet18) achieving 51-116x speedups, and complex generative models (CycleGAN) reaching 11x improvements compared to CPU training. Additionally, in our PyTorch vs. TensorFlow comparison, we observed that TensorFlow's kernel-fusion optimizations reduce inference latency by approximately 15%. We also analyze GPU memory usage trends and projecting requirements through 2025 using polynomial regression. Our findings highlight that while GPUs are essential for sustaining AI's growth, democratized and shared access to GPU resources is critical for enabling research innovation across institutions with limited computational budgets.
toXiv_bot_toot
@arXiv_mathOA_bot@mastoxiv.page2026-03-24 08:13:37
Polynomials in $c$-free random variables with applications to free denoising
Adrian Celestino, Franz Lehner, Kamil Szpojankowski
https://arxiv.org/abs/2603.21372 https://arxiv.org/pdf/2603.21372 https://arxiv.org/html/2603.21372
arXiv:2603.21372v1 Announce Type: new
Abstract: We study distributions of polynomials in conditionally free (c-free) random variables, a notion of independence for two-state noncommutative probability spaces introduced by Bozejko, Leinert and Speicher. To this end we establish recursive relations between the joint Boolean cumulants of c-free random variables, analogous to previously found recursions for Boolean cumulants of free random variables. The algebraic reformulation of these recursions on the free associative algebra provides an effective formal machinery for the computation of the moment generating functions and thus the distributions of arbitrary self-adjoint polynomials in c-free random variables. As an application of a recent observation, our approach can be used to determine conditional expectations of the form $E[a|P(a,b)]$, where $P(a,b)$ is a self-adjoint polynomial in free (in the sense of Voiculescu) random variables $a,b$. We illustrate this with an example where $P(a,b)=i[a,b]$. Finally we define orthogonal projections that formally play the role of conditional expectations in the framework of c-freeness and share some properties with the conditional expectations of free variables. In particular they can be used to re-derive by purely algebraic methods the formula of Popa and Wang for the $\Sigma$-transform for the c-free multiplicative convolution.
toXiv_bot_toot
@nfdi4culture@nfdi.social2026-03-20 14:22:55
🎉Hallo, liebe 4Culture-Community! Unsere Kolleg:innen waren diese Woche (16.–20.03.2026) in Poznań, um an der zweiten ECHOES-Jahresveranstaltung teilzunehmen! Das Treffen brachte alle ECHOES-Partner, Schwesterprojekte und die ECCCH-Community zusammen, um der Europäischen Kommission über den aktuellen Stand,, die bisherigen Ergebnisse und Aktivitäten für die kommenden Monate zu berichten:
@cyclocracy@norden.social2026-02-22 13:35:01
@thijs_lucas@norden.social2026-02-16 15:36:20
Bürokratieabbau ist kein Problem. Umweltschutzauflagen sind kein Problem - sie lassen sich planen, vorbereiten und sauber in Projekte eintakten.
Politik mit difuser Haltung und Entscheidungsschwächen sind ein schwer planbarer Grund für Projektverzug und eine Herausforderung für #Planungsbeschleunigung
Übrigens nicht nur in Berlin mit der CDU, wie hier von
@khalidabuhakmeh@mastodon.social2026-04-16 18:29:34
This quote from Intuit about Anthropic Claude Opus 4.7 is amazing for its non-committal tone. The tap-dancing in this single quote would make Fred Astaire jealous. 😅
@heiseonline@social.heise.de2026-03-06 12:52:00