Your personalresearch library

Discover papers, track what you're reading, rate and review.Build a library that grows with your curiosity.

Get Started Explore Papers

Currently Reading

Want to Read

All Papers

Advanced filtersTopic: All

Experimental

Topic

A [CII]-158-micron Survey of DLA galaxies at z~4: Observations of Dense and Metal-Enriched Neutral Gas within the Circumgalactic Medium of Star-Forming Galaxies

Marcel Neeleman, Nissim Kanekar, J. Xavier Prochaska, Marc A. Rafelski, Lordrick A. Kahinga

arXiv·2025

We present a survey undertaken with the Atacama Large Millimeter/submillimeter Array (ALMA) to study the galaxies associated with a representative sample of 16 damped Ly-alpha absorbers (DLAs) at z~4.1-4.5, using the [CII]-158-micron ([CII]) line. We detect seven [CII]-emitting galaxies in the fields of 5 DLAs, all of which have absorption metallicity [M/H] > -1.5. We find that the detectability of these HI-selected galaxies with ALMA is a strong function of DLA metallicity, with a detection rate of 71^{+11}_{-20}% for DLAs with [M/H] > -1.5 and 0^{+18}% for DLAs with [M/H] <-1.5. The identified DLA galaxies have far-infrared properties similar to those of typical star-forming galaxies at z ~ 4, with estimated obscured star-formation rates ranging from ~6 Msun/yr to 110 Msun/yr. High-metallicity DLAs therefore provide an efficient way to identify and study samples of high-redshift, star-forming galaxies without preselecting the galaxies by their emission properties. The agreement between the velocities of the metal absorption lines of the DLA and the [CII] emission line of the DLA galaxy indicates that the metals within the DLA originated in the galaxy. With observed impact parameters between 14 and 59 kpc, this indicates that star-forming galaxies at z ~ 4 have a substantial reservoir of dense, cold neutral gas within their circumgalactic medium that has been enriched with metals from the galaxy.

No ratings yet

View paper →

A Polynomial-Time Algorithm for Variational Inequalities under the Minty Condition

Ioannis Anagnostides, Gabriele Farina, Tuomas Sandholm, Brian Hu Zhang

arXiv·2025

Solving variational inequalities (SVIs) is a foundational problem at the heart of optimization. However, this expressivity comes at the cost of computational hardness. As a result, most research has focused on carving out specific subclasses that elude those intractability barriers. A classical property that goes back to the 1960s is the Minty condition, which postulates that the Minty VI (MVI) problem admits a solution. In this paper, we establish the first polynomial-time algorithm -- that is, with complexity growing polynomially in the dimension $d$ and $\log(1/ε)$ -- for solving $ε$-SVIs for Lipschitz continuous mappings under the Minty condition. Prior approaches either incurred an exponentially worse dependence on $1/ε$ or made restrictive assumptions. To do so, we introduce a new variant of the ellipsoid algorithm whereby separating hyperplanes are obtained after taking a gradient descent step from the center of the ellipsoid. It succeeds even though the set of SVIs can be nonconvex and not fully dimensional. Moreover, when our algorithm is applied to an instance with no MVI solution and fails to identify an SVI solution, it produces a succinct certificate of MVI infeasibility. We also show that deciding whether the Minty condition holds is $\mathsf{coNP}$-complete, thereby establishing that the disjunction of those two problems is polynomial-time solvable even though each problem is individually intractable. We provide several extensions and new applications of our main results. Most notably, we obtain the first polynomial-time algorithms for i) globally minimizing a (potentially nonsmooth) quasar-convex function, and ii) computing Nash equilibria in multi-player harmonic games. Finally, in two-player general-sum concave games, we give the first polynomial-time algorithm that outputs either a Nash equilibrium or a strict coarse correlated equilibrium.

No ratings yet

View paper →

PreviousPage 173 of 613Next

Sample, Don't Search: Rethinking Test-Time Alignment for Language Models

Gonçalo Faria, Noah A. Smith

arXiv·2025

Increasing test-time computation has emerged as a promising direction for improving language model performance, particularly in scenarios where model finetuning is impractical or impossible due to computational constraints or private model weights. However, existing test-time search methods using a reward model (RM) often degrade in quality as compute scales, due to the over-optimization of what are inherently imperfect reward proxies. We introduce QAlign, a new test-time alignment approach. As we scale test-time compute, QAlign converges to sampling from the optimal aligned distribution for each individual prompt. By adopting recent advances in Markov chain Monte Carlo for text generation, our method enables better-aligned outputs without modifying the underlying model or even requiring logit access. We demonstrate the effectiveness of QAlign on mathematical reasoning benchmarks (GSM8K and GSM-Symbolic) using a task-specific RM, showing consistent improvements over existing test-time compute methods like best-of-n and majority voting. Furthermore, when applied with more realistic RMs trained on the Tulu 3 preference dataset, QAlign outperforms direct preference optimization (DPO), best-of-n, majority voting, and weighted majority voting on a diverse range of datasets (GSM8K, MATH500, IFEval, MMLU-Redux, and TruthfulQA). A practical solution to aligning language models at test time using additional computation without degradation, our approach expands the limits of the capability that can be obtained from off-the-shelf language models without further training.

No ratings yet

View paper →

Real-World Data Confirm That the Integration of Deuterium Depletion into Conventional Cancer Therapy Multiplies the Survival Probability of Patients

Gábor Somlyai, András Papp, Ildikó Somlyai, Beáta Zs Kovács, Mária Debrődi

Biomedicines·2025·1 citations

Background: Over thirty years of basic research has demonstrated that the deuterium-to-hydrogen ratio plays a pivotal role in regulating metabolism and cell growth via a sub-molecular regulatory system that orchestrates the intricate complexity of life in eukaryotic organisms. Deuterium depletion, achieved through deuterium-depleted water (DDW), has shown anticancer effects in vitro, in vivo, and in Phase 2 prospective and retrospective clinical studies. Methods: In this population-based observational study, 2649 cancer patients undergoing conventional therapy and consuming DDW were included between October 1992 and October 2024. With various cancer types and stages and conventional therapies received, they are representing a broad spectrum of the Hungarian cancer population. Survival was selected as the primary endpoint, and the median survival time (MST) of these patients and various subgroups was calculated and compared to the overall Hungarian cancer population’s MST of 2.4 years. Results: For the entire study population, MST from diagnosis was 12.4 years (95% CI: 9.8–14.9), and from the initiation of DDW treatment, 7.6 years (95% CI: 5.9–9.3). Conclusions: Utilizing DDW enables targeted intervention in the sub-molecular regulatory system, paving the way for innovative therapeutic applications and a more profound understanding of cellular processes. Integrating deuterium depletion into conventional cancer therapies has the potential to significantly enhance survival rates and reduce cancer-related mortality by 75–80%.

★ 2.0 (1)

View paper →

Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning

Julian Minder, Clément Dumas, Caden Juang, Bilal Chugtai, Neel Nanda

arXiv·2025

Model diffing is the study of how fine-tuning changes a model's representations and internal algorithms. Many behaviors of interest are introduced during fine-tuning, and model diffing offers a promising lens to interpret such behaviors. Crosscoders are a recent model diffing method that learns a shared dictionary of interpretable concepts represented as latent directions in both the base and fine-tuned models, allowing us to track how concepts shift or emerge during fine-tuning. Notably, prior work has observed concepts with no direction in the base model, and it was hypothesized that these model-specific latents were concepts introduced during fine-tuning. However, we identify two issues which stem from the crosscoders L1 training loss that can misattribute concepts as unique to the fine-tuned model, when they really exist in both models. We develop Latent Scaling to flag these issues by more accurately measuring each latent's presence across models. In experiments comparing Gemma 2 2B base and chat models, we observe that the standard crosscoder suffers heavily from these issues. Building on these insights, we train a crosscoder with BatchTopK loss and show that it substantially mitigates these issues, finding more genuinely chat-specific and highly interpretable concepts. We recommend practitioners adopt similar techniques. Using the BatchTopK crosscoder, we successfully identify a set of chat-specific latents that are both interpretable and causally effective, representing concepts such as $\textit{false information}$ and $\textit{personal question}$, along with multiple refusal-related latents that show nuanced preferences for different refusal triggers. Overall, our work advances best practices for the crosscoder-based methodology for model diffing and demonstrates that it can provide concrete insights into how chat-tuning modifies model behavior.

No ratings yet

View paper →

Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasets

Chuning Zhu, Raymond Yu, Siyuan Feng, Benjamin Burchfiel, Paarth Shah, Abhishek Gupta

arXiv·2025

Imitation learning has emerged as a promising approach towards building generalist robots. However, scaling imitation learning for large robot foundation models remains challenging due to its reliance on high-quality expert demonstrations. Meanwhile, large amounts of video data depicting a wide range of environments and diverse behaviors are readily available. This data provides a rich source of information about real-world dynamics and agent-environment interactions. Leveraging this data directly for imitation learning, however, has proven difficult due to the lack of action annotation. In this work, we present Unified World Models (UWM), a framework that allows for leveraging both video and action data for policy learning. Specifically, a UWM integrates an action diffusion process and a video diffusion process within a unified transformer architecture, where independent diffusion timesteps govern each modality. By controlling each diffusion timestep, UWM can flexibly represent a policy, a forward dynamics, an inverse dynamics, and a video generator. Through simulated and real-world experiments, we show that: (1) UWM enables effective pretraining on large-scale multitask robot datasets with both dynamics and action predictions, resulting in more generalizable and robust policies than imitation learning, (2) UWM naturally facilitates learning from action-free video data through independent control of modality-specific diffusion timesteps, further improving the performance of finetuned policies. Our results suggest that UWM offers a promising step toward harnessing large, heterogeneous datasets for scalable robot learning, and provides a simple unification between the often disparate paradigms of imitation learning and world modeling. Videos and code are available at https://weirdlabuw.github.io/uwm/.

No ratings yet

View paper →

Longitudinal Association Between the Consumption of Vegetables, Fruits, and Red Meat and Diabetes Disease Burden: An Analysis of Multiple Global Datasets

Manqiong Yuan, Juan Wang, Lifen Jin, Liangwen Zhang, Ya Fang

Nutrients·2025·1 citations

Background: Dietary factors, such as vegetable, fruit, and red meat consumption, have varying effects on the disease burden of diabetes, the world’s third leading health concern. This study aims to evaluate the global impact of vegetable/fruit/red meat consumption on disease burdens. Methods: Diabetes disease burden, vegetable/fruit/red meat consumption, and covariates data were obtained from the Global Burden of Disease Study (GBD) 2021, Food and Agriculture Organization (FAO), and WHO, respectively, and matched by country/region and year. Global vector maps assessed the status of diabetes disease burden and the consumption of three dietary factors in 2021, and their trends from 2010 to 2021 were illustrated through local regression curves. Generalized additive mixed models (GAMMs) were used to analyze relationships, with weights assigned based on log-transformed values relative to the mean population of each country. Results: A comprehensive dataset spanning 2010–2021, encompassing 175 countries/regions, was successfully matched and utilized in the analysis. In 2021, Oceania had the highest diabetes burden, whereas East Asia had a lower one. Globally, the disease burden increased from 2010 to 2021, accompanied by rising per capita vegetable and fruit consumption but declining red meat consumption. Vegetable consumption was inversely correlated with the age-standardized incidence rate (ASIR) and exhibited a “J-shaped” curve with the age-standardized mortality rate (ASMR) and age-standardized disability-adjusted life year (DALY) rate (ASDR) (nadir at approximately 80 kcal/capita/day (kcal/cap/day). Fruit consumption had a “U-shaped” relationship with ASDR (nadir at approximately 100 kcal/cap/day). Red meat consumption was negatively correlated to ASIR and had a “U-shaped” relationship with ASMR and ASDR (nadir at 200 kcal/cap/day). Conclusions: The global diabetes disease burden is heavy, and dietary consumption varies widely. Vegetable-related risks differ between diabetics and non-diabetics. Proper fruit consumption decreases ASDR. Moderate red meat increases can reduce the disease burden, but excessive consumption increases ASMR and ASDR.

★ 1.0 (1)

View paper →

Quartz Crystal Fishtail Projectile Point and Paiján Technology from the Pampa Lechuza Site, South-Central Coastal Desert of Peru

Antonio Pérez-Balarezo, Jalh Dulanto, Marcellus de Almeida, Jesús Briceño, Angelo Constantine, Enrique Guadalupe, Edwin Silva, Abel Icochea, Daniel Vergara, Jessica Riquez

PaleoAmerica·2025

Fishtail projectile points, found from Central America to southern South America during the Late Pleistocene, are typically flint-based but occasionally crafted from quartz. This work examines quartz crystal specimens from Peru’s Pampa Lechuza site (Ica Valley), dated to the Late Pleistocene–Early Holocene, including Fishtail and Paiján Points. The co-occurrence of these technologies in south-central Peru suggests a more intricate scenario of Andean peopling during the Pleistocene–Holocene transition.

No ratings yet

View paper →

Your personalresearch library

All Papers

A [CII]-158-micron Survey of DLA galaxies at z~4: Observations of Dense and Metal-Enriched Neutral Gas within the Circumgalactic Medium of Star-Forming Galaxies

A Polynomial-Time Algorithm for Variational Inequalities under the Minty Condition

Sample, Don't Search: Rethinking Test-Time Alignment for Language Models

Real-World Data Confirm That the Integration of Deuterium Depletion into Conventional Cancer Therapy Multiplies the Survival Probability of Patients

The Hereditary Secret No One Else Noticed: Aspen Trees | Thought Catalog

Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning

Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasets

Slow Read Issue 1: Messy Truths

Longitudinal Association Between the Consumption of Vegetables, Fruits, and Red Meat and Diabetes Disease Burden: An Analysis of Multiple Global Datasets

You're Not 'Too Controlling' You Just Haven't Learned to Use It to You

Quartz Crystal Fishtail Projectile Point and Paiján Technology from the Pampa Lechuza Site, South-Central Coastal Desert of Peru

The Israel Exemption