Oldest Papers

Date: Oldest Clear all

Advanced filtersTopic: All

Experimental

Topic

Topics are auto-detected from title, abstract, and metadata and may be imperfect.

Publication Date

Newest Oldest Clear

Average Rating

Highest Lowest

Complex Systems

Saffron Huang

The Point Magazine·2025

What does it mean to measure intelligence?

No ratings yet

View paper →

The nature of the relation between mental well-being and ill-being

Christian K. Tamnes, Mona Bekkhus, Maja Eilertsen, Ragnhild B. Nes, Monica Beer Prydz, Eivind Ystrom, Eira R. Aksnes, Synøve N. Andersen, Helga Ask, Ziada Ayorech, Tina Baier, Dani Beck, Eirik Jerven Berger, Ludvig D. Bjørndal, Olga D. Boer, Marieke G. N. Bos, Avshalom Caspi, Rosa Cheesman, Razieh Chegeni, Laura Cortés-García, Perline A. Demange, Joakim C. Ebeltoft, Maud Edvoll, Nikolai Haahjem Eftedal, Espen Moen Eilertsen, Yuan Fang, Lia Ferschmann, Sam Fluit, Siri H. Folstad, Eivor Fredriksen, Liridona Gashi, Kristin Gustavson, Thomas Hansen, Live S. Hofgaard, Valerie Karl, Michal Kozák, Kristian G. Krogshus, Jinrui Liu, Torkild H. Lyngstad, Niamh MacSweeney, Hermine H. M. Maes, Lilian Mayerhofer, Tom A. McAdams, Terrie E. Moffitt, Nasimeh Naseri, Thomas S. Nilsen, Linn B. Norbom, Egil Nygaard, Aurora Oftedal, Willy Pedersen, Qi Qin, Rubén Rodríguez-Cano, Vegard Skirbekk, Dinka Smajlagic, Ole Martin Solberg, Anneli D. Tandberg, Irene J. E. Teulings, Fartein Ask Torvik, Stella Tsotsi, Vidar S. Ulset, Emily G. Vira, Espen Røysamb, Tilmann von Soest

Nature Human Behaviour·2025·2 citations

No abstract available

No ratings yet

View paper →

To Infinity and Beyond: Tool-Use Unlocks Length Generalization in State Space Models

Eran Malach, Omid Saremi, Sinead Williamson, Arwen Bradley, Aryo Lotfi, Emmanuel Abbe, Josh Susskind, Etai Littwin

arXiv·2025

State Space Models (SSMs) have become the leading alternative to Transformers for sequence modeling. Their primary advantage is efficiency in long-context and long-form generation, enabled by fixed-size memory and linear scaling of computational complexity. We begin this work by showing a simple theoretical result stating that SSMs cannot accurately solve any ``truly long-form'' generation problem (in a sense we formally define), undermining their main competitive advantage. However, we show that this limitation can be mitigated by allowing SSMs interactive access to external tools. In fact, we show that given the right choice of tool access and problem-dependent training data, SSMs can learn to solve any tractable problem and generalize to arbitrary problem length/complexity (i.e., achieve length generalization). Following our theoretical finding, we demonstrate that tool-augmented SSMs achieve remarkable length generalization on a variety of arithmetic, reasoning, and coding tasks. These findings highlight SSMs as a potential efficient alternative to Transformers in interactive tool-based and agentic settings.

No ratings yet

View paper →

PreviousPage 363 of 505Next

The Launching of Galactic Winds from a Multiphase ISM

Fernando Hidalgo-Pineda, Max Gronke, Philipp Grete

arXiv·2025

Galactic outflows are a key agent of galaxy evolution, yet their observed multiphase nature remains difficult to reconcile with theoretical models, which often fail to explain how cold gas survives interactions with hot, fast winds. We present high-resolution 3D hydrodynamic simulations of hot outflows interacting with a multiphase interstellar medium (ISM), parameterised by its cold-gas volume filling fraction $f_v$, depth $L_{\rm ISM}$, and clump size $r_{\rm cl}$. We identify a universal survival criterion $f_v L_{\rm ISM} \gtrsim r_{\rm crit}$ that generalises the classical single-cloud condition ($r_{\rm cl} > r_{\rm crit}$) and correctly predicts cold-gas survival across diverse ISM configurations - including scale-free - down to $r_{\rm cl}/r_{\rm crit} \sim 10^{-2}$. The surviving cold phase rapidly loses memory of the initial ISM structure and evolves toward a self-similar clump mass spectrum following Zipf's law ($\mathrm{d}N/\mathrm{d}m \propto m^{-2}$), implying that turbulent mixing and radiative condensation universally shape multiphase outflows. Cold gas assembles into plumes or confined shells of size $\sim χr_{\mathrm{cl,min}}$, growing as mass is accreted from the hot phase. The interaction of a laminar wind with a clumpy ISM drives turbulence in both phases, with first-order velocity structure functions following a Kolmogorov scaling and an injection scale set by $L_{\rm ISM}$, while velocity dispersions reach $σ\sim c_{\rm s,cold}$. The areal covering fraction of cold gas approaches unity even for $f_v \sim 10^{-3}$, though its volume filling fraction stays low, explaining the "misty" appearance of observed outflows. Together, these results link small-scale cloud-wind interactions to galaxy-scale feedback, and we discuss their implications for interpreting observations and for modelling multiphase galactic winds in larger-scale simulations.

No ratings yet

View paper →

RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning

Kun Lei, Huanyu Li, Dongjie Yu, Zhenyu Wei, Lingxiao Guo, Zhennan Jiang, Ziyu Wang, Shiyu Liang, Huazhe Xu

arXiv·2025

Real-world robotic manipulation in homes and factories demands reliability, efficiency, and robustness that approach or surpass the performance of skilled human operators. We present RL-100, a real-world reinforcement learning framework built on diffusion-based visuomotor policies. RL-100 unifies imitation and reinforcement learning under a single PPO-style objective applied within the denoising process, yielding conservative and stable policy improvements across both offline and online stages. To meet deployment latency constraints, we employ a lightweight consistency distillation procedure that compresses multi-step diffusion into a one-step controller for high-frequency control. The framework is task-, embodiment-, and representation-agnostic, and supports both single-action outputs and action-chunking control. We evaluate RL-100 on seven diverse real-robot manipulation tasks, ranging from dynamic pushing and agile bowling to pouring, cloth folding, unscrewing, and multi-stage juicing. RL-100 attains 100% success across evaluated trials, achieving 900 out of 900 successful episodes, including up to 250 out of 250 consecutive trials on one task, and matches or surpasses expert teleoperators in time-to-completion. Without retraining, a single policy attains approximately 90% zero-shot success under environmental and dynamics shifts, adapts in a few-shot regime to significant task variations (86.7%), and remains robust to aggressive human perturbations (about 95%). In a public shopping-mall deployment, the juicing robot served random customers continuously for roughly seven hours without failure. Together, these results suggest a practical path toward deployment-ready robot learning: start from human priors, align training objectives with human-grounded metrics, and reliably extend performance beyond human demonstrations.

No ratings yet

View paper →

Reasoning with Sampling: Your Base Model is Smarter Than You Think

Aayush Karan, Yilun Du

arXiv·2025

Frontier reasoning models have exhibited incredible capabilities across a wide array of disciplines, driven by posttraining large language models (LLMs) with reinforcement learning (RL). However, despite the widespread success of this paradigm, much of the literature has been devoted to disentangling truly novel behaviors that emerge during RL but are not present in the base models. In our work, we approach this question from a different angle, instead asking whether comparable reasoning capabilites can be elicited from base models at inference time by pure sampling, without any additional training. Inspired by Markov chain Monte Carlo (MCMC) techniques for sampling from sharpened distributions, we propose a simple iterative sampling algorithm leveraging the base models' own likelihoods. Over different base models, we show that our algorithm offers substantial boosts in reasoning that nearly match and even outperform those from RL on a wide variety of single-shot tasks, including MATH500, HumanEval, and GPQA. Moreover, our sampler avoids the collapse in diversity over multiple samples that is characteristic of RL-posttraining. Crucially, our method does not require training, curated datasets, or a verifier, suggesting broad applicability beyond easily verifiable domains.

No ratings yet

View paper →

VT-Refine: Learning Bimanual Assembly with Visuo-Tactile Feedback via Simulation Fine-Tuning

Binghao Huang, Jie Xu, Iretiayo Akinola, Wei Yang, Balakumar Sundaralingam, Rowland O'Flaherty, Dieter Fox, Xiaolong Wang, Arsalan Mousavian, Yu-Wei Chao, Yunzhu Li

arXiv·2025

Humans excel at bimanual assembly tasks by adapting to rich tactile feedback -- a capability that remains difficult to replicate in robots through behavioral cloning alone, due to the suboptimality and limited diversity of human demonstrations. In this work, we present VT-Refine, a visuo-tactile policy learning framework that combines real-world demonstrations, high-fidelity tactile simulation, and reinforcement learning to tackle precise, contact-rich bimanual assembly. We begin by training a diffusion policy on a small set of demonstrations using synchronized visual and tactile inputs. This policy is then transferred to a simulated digital twin equipped with simulated tactile sensors and further refined via large-scale reinforcement learning to enhance robustness and generalization. To enable accurate sim-to-real transfer, we leverage high-resolution piezoresistive tactile sensors that provide normal force signals and can be realistically modeled in parallel using GPU-accelerated simulation. Experimental results show that VT-Refine improves assembly performance in both simulation and the real world by increasing data diversity and enabling more effective policy fine-tuning. Our project page is available at https://binghao-huang.github.io/vt_refine/.

No ratings yet

View paper →

Continual Learning via Sparse Memory Finetuning

Jessy Lin, Luke Zettlemoyer, Gargi Ghosh, Wen-Tau Yih, Aram Markosyan, Vincent-Pierre Berges, Barlas Oğuz

arXiv·2025

Modern language models are powerful, but typically static after deployment. A major obstacle to building models that continually learn over time is catastrophic forgetting, where updating on new data erases previously acquired capabilities. Motivated by the intuition that mitigating forgetting is challenging because trainable parameters are shared across all tasks, we investigate whether sparse parameter updates can enable learning without catastrophic forgetting. We introduce sparse memory finetuning, leveraging memory layer models (Berges et al., 2024), which are sparsely updated by design. By updating only the memory slots that are highly activated by a new piece of knowledge relative to usage on pretraining data, we reduce interference between new knowledge and the model's existing capabilities. We evaluate learning and forgetting compared to full finetuning and parameter-efficient finetuning with LoRA on two question answering tasks. We find that sparse memory finetuning learns new knowledge while exhibiting substantially less forgetting: while NaturalQuestions F1 drops by 89% after full finetuning on new facts and 71% with LoRA, sparse memory finetuning yields only an 11% drop with the same level of new knowledge acquisition. Our results suggest sparsity in memory layers offers a promising path toward continual learning in large language models.

No ratings yet

View paper →

What is the True HI Gas Content in Massive Quiescent Galaxies in the Local Universe?

Xiao Li, Cheng Li, Cheng Cheng, Houjun Mo, Jing Wang, Amélie Saintonge

arXiv·2025

While massive quiescent galaxies are known to be poor in atomic hydrogen (HI), their true HI content remains poorly constrained due to the limited sensitivity and morphological biases of existing surveys. We present deep HI observations using the Five-hundred-meter Aperture Spherical radio Telescope (FAST) for a representative sample of 78 low-redshift massive quiescent galaxies, selected by stellar mass ($M_\ast > 10^{10} M_\odot$) and color (NUV$-r > 5$). Our observations reach an exceptional detection limit of $\log(M_{HI}/M_\ast) = -3$. We find that one-third of the sample follows the predicted HI mass fraction distribution based on previous surveys. However, the majority ($\sim$2/3) exhibit significantly lower HI content, with a $3σ$ upper limit of $\log(M_{HI}/M_\ast) < -3.46$ derived from stacking the non-detections. As a consequence of this enormous dynamic range and the high fraction of non-detections, the HI mass fraction shows no clear correlation with parameters tracing star formation, structure, or morphology. Our FAST sample shows remarkable similarity to the $ATLAS^{3D}$ sample which only includes early-type galaxies, both in its high fraction of HI-poor galaxies and its high satellite fraction among HI-poor galaxies, as indicated by three different environmental characteristics: projected cross-correlation functions, background-subtracted neighbour counts, and central/satellite classification in the SDSS galaxy group catalog. These results suggest that while early-type morphology and environment play crucial roles in suppressing HI reservoirs, neither factor alone fully explains the observed gas depletion, indicating that additional physical mechanisms must be responsible for the extreme HI deficiency prevalent in massive quiescent galaxies.

No ratings yet

View paper →

An extraordinary colonial spider community in Sulfur Cave (Albania/Greece) sustained by chemoautotrophy

István Urák, Blerina Vrenozi, Zofia Głąbiak, Ninon Lecoquierre, Cord Eiberger, Mark Maraun, Andrei Ştefan, Jean-François Flot, Traian Brad, Luisa Dainelli, Serban M. Sarbu, Raluca I. Băncilă

Subterranean Biology·2025

We report the discovery and detailed analysis of an extraordinary colonial spider assemblage in Sulfur Cave, a chemoautotrophic sulfidic ecosystem located on the Albania-Greece border. The colony, comprising an estimated 69,000 individuals of Tegenaria domestica (Agelenidae) and more than 42,000 of Prinerigone vagans (Linyphiidae), spans a surface area of over 100 m²—representing the first documented case of colonial web formation in these species. Stable isotope analyses (δ¹³C and δ¹⁵N) revealed that the trophic web sustaining this assemblage is fueled by in situ primary production from sulfur-oxidizing microbial biofilms then transferred through chironomid larvae and adults to higher trophic levels. Morphological and molecular data confirmed the identity of the two spider species and revealed that their populations in Sulfur Cave are genetically distinct from other populations. Regarding T. domestica, we found a seasonal pattern in fecundity, with significantly larger egg clutches in early summer. Microbiome analysis of this species also revealed a lower Shannon diversity in the cave population compared with a surface individual captured nearby. Our findings unveil a unique case of facultative coloniality in this cosmopolitan spider, likely driven by resource abundance in a chemoautotrophic cave, and provide new insights into the adaptation and trophic integration of surface species in sulfidic subterranean habitats.

No ratings yet

View paper →

Are cooperatives more virtuous than corporations?

Joseph Heath

Politics, Philosophy & Economics·2025

There are several sectors of the economy in which cooperatives have flourished, competing successfully against standard business corporations. The best explanation for their success is that they provide superior benefits to their members . The question addressed by this paper is whether cooperatives also provide important benefits to society, such that non-members should prefer a cooperative economy to one dominated by business corporations. It has often been suggested that cooperatives are more virtuous because they are more democratic, less hierarchical, less anti-social and less apt to produce economic inequality. This paper evaluates these claims. The central challenge stems from the observation that cooperatives are not nearly as different from corporations as is commonly assumed.

★ 4.0 (1)

View paper →

Oldest Papers

Complex Systems

The nature of the relation between mental well-being and ill-being

To Infinity and Beyond: Tool-Use Unlocks Length Generalization in State Space Models

The Launching of Galactic Winds from a Multiphase ISM

RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning

Reasoning with Sampling: Your Base Model is Smarter Than You Think

VT-Refine: Learning Bimanual Assembly with Visuo-Tactile Feedback via Simulation Fine-Tuning

Continual Learning via Sparse Memory Finetuning

What is the True HI Gas Content in Massive Quiescent Galaxies in the Local Universe?

An extraordinary colonial spider community in Sulfur Cave (Albania/Greece) sustained by chemoautotrophy

Are cooperatives more virtuous than corporations?

Design for Hope: Cultivating Deliberate Hope in the Face of Complex Societal Challenges