Pan and Yang study stochastic nonconvex–strongly-convex bilevel optimization when only unbiased stochastic gradient estimates are available. They propose MRT-FD, a single-loop algorithm that simultaneously tracks the upper-level variable, lower-level solution, and an auxiliary response variable, using order-$p$ finite differences to reconstruct hypergradient actions from first-order queries. For lower-level objectives with order-$p$ smoothness, the method achieves $\mathcal{O}(\varepsilon^{-4-2/p})$ oracle complexity, and the authors prove a matching $\Omega(\varepsilon^{-4-2/p})$ lower bound, settling the optimal $\varepsilon$-dependence for any fixed finite $p$.

Varda-single-1.0 is a deterministic, data-driven weather forecasting system built for the Alpine domain, delivering hourly forecasts at 1 km resolution via a stretched-grid Graph Transformer architecture. It combines a 6-hourly autoregressive forecaster with a temporal downscaler to balance forecast horizon and temporal fidelity. The work demonstrates that machine learning models can match or exceed operational numerical weather prediction baselines in complex topography, and it openly releases model weights and software to enable operational adoption and downstream research.

This paper asks whether deep self-attention networks can exactly interpolate arbitrary finite datasets using only depth, without adapting parameters to the task. The authors prove that universal interpolation is possible with as few as two frozen Gaussian-initialized single-head residual attention blocks, composed with data-dependent signs, durations, and ordering. They also characterize when causal masking prevents interpolation. The results matter because they provide a theoretical foundation for looped and recurrent-depth transformers that share parameters across layers.

STORM is a single-sample stochastic recursive momentum method that avoids periodic full-gradient checkpoints. Prior analyses required a stronger average smoothness assumption and produced suboptimal or logarithmically degraded rates. This paper shows that the original STORM recursion alone achieves the optimal $O(T^{-1/4})$ nonconvex rate, $O(\sigma R/\sqrt{T})$ convex rate, and $O(\sigma^2/(\lambda T))$ strongly convex rate under standard smoothness, by introducing auxiliary sequences and shifted iterates to cancel the estimator bias.

Judgment models that return option probabilities without reasoning text are attractive as fast agent policies, but their competence boundary is unclear. This paper argues that such models excel at evaluation (judging given options against described criteria) yet fail at simulation (predicting missing consequences like opponent moves or physical dynamics). By cleanly separating the two, the authors show that supplying simulated outcomes via code restores strong performance across matrix games, text adventures, and robot control. The result is a sharp diagnostic of where fast judgment models can be trusted and a principled architecture for combining them with external simulators.

Asynchronous LLM post-training mixes stale rollouts from earlier policies, making GRPO-style optimization unstable. This paper derives a convergence bound that exposes a second-moment versus bias tradeoff, then proposes Group-Mass Capping GRPO (GMC-GRPO), which rescales groups when their average importance ratio exceeds a cap. The result is tighter delay dependence than per-trajectory clipping and stronger empirical robustness under large rollout delays on Qwen3 models.

This paper dissects the group attention mechanism in the time series foundation model Chronos-2, which is advertised to serve both multivariate (MV) and in-context learning (ICL) forecasting. The authors decompose attention into its value/output ($V$/$O$) pooling pathway and its query/key ($Q$/$K$) learned weighting pathway by editing the attention matrix $\alpha$ at inference time. Their core finding is that uniform pooling helps nearly all tested configurations, while learned $Q$/$K$ weighting helps MV forecasting but systematically degrades ICL—sometimes leaving forecasts worse than univariate inference. The work matters because it identifies a cheap, retraining-free fix: forcing $\alpha$ uniform in block 0 alone fully recovers ICL performance across every configuration tested.

This paper asks how continuous latent reasoning in chemistry can be improved by training thoughts to anticipate future outcomes without verbalizing every intermediate step. The authors introduce Latent-JEPA, which augments a recurrent latent reasoner with joint-embedding prediction of both textual continuations and molecular structures using pretrained encoders. Evaluated on ChemCoTBench, the framework shows the strongest task gains in molecular optimization and yields latent representations that are more chemically structured.

This paper tackles validation congestion in scientific discovery, where enormous virtual libraries overwhelm scarce experimental capacity. Instead of relying on data-hungry regressors, it proposes PRISMS, a framework that curates candidates using multi-fidelity pairwise rankings aggregated through a regularized Bradley-Terry model, escalating queries to expensive experts via a Fisher-information criterion. The central claim is that comparative judgments from heterogeneous experts can slash the number of slow experimental rounds needed to find top-performing designs.

Selecting the best large language model for a live stream of prompts is difficult because benchmark scores are only coarse proxies for real-world performance and oracle annotations are costly. This paper proposes Online LLM Picker, the first active model selection framework designed specifically for LLMs in streaming settings: it uses an Exponential Weights posterior and an adaptive query probability based on hypothetical response variance scaled by posterior entropy to decide which prompts are worth annotating. Evaluated across over 130 models and 10 datasets spanning question answering, dialogue, grammar correction, and math, the method claims to cut annotation cost by up to 71.67% and reduce regret by up to 2.51× compared to baselines.

Continual learners that keep a task-specific adapter in every ViT block accumulate parameters linearly in the number of tasks. This paper asks which blocks should specialize, showing that exhaustive placement search traces an inverted-U curve peaking at intermediate depth, while cheap internal criteria systematically favor the deep but suboptimal layers. To sidestep expensive search, the authors propose LS-B, which observes the first few tasks through a frozen fMRI encoding model of human visual cortex and commits task-specific capacity to the blocks whose brain readouts vary most across tasks relative to their stable structure.

Reinforcement learning for reachability specifications has been limited to model-based methods that explicitly estimate transition probabilities. This paper introduces Quasar, a model-free Q-learning variant that provably converges to optimal reachability policies on MDPs without non-terminal maximal end components, using only $O(|S||A|)$ memory. By replacing the missing discount factor with an intrinsic expected hitting time, the authors obtain a weighted contraction and demonstrate orders-of-magnitude sample improvements over the state of the art.

Iterative tractography must reconcile noisy local diffusion signals with globally plausible fiber pathways, yet standard deep learning models optimize only a local next-step loss. This paper tackles that gap by systematically evaluating RNN and Transformer architectures and introducing a generation-validation (GV) phase that supervises training at the streamline level via connectivity scores. It reports top scores on the ISMRM 2015 phantom and provides open-source models, while also arguing that current phantoms and in vivo evaluation tools are too limited for definitive benchmarking.

Continual reinforcement learning demands a delicate balance between plasticity (learning new tasks) and stability (not forgetting old ones). This paper argues that neuroevolution—specifically evolution strategies (ES) and genetic algorithms (GAs)—offers a compelling alternative to gradient-based policy-gradient methods, which tend to lose plasticity as tasks change. Across a broad sweep of environments and task-shift types, the authors show that ES consistently achieves the best stability-plasticity trade-off, and they propose a clear mechanism: parameter-space exploration produces wider "neighborhoods" of robust solutions that overlap across sequential tasks. The work matters because it challenges the prevailing focus on fixing gradient-based RL and instead revives a classic optimization paradigm with strong empirical and conceptual justification.

This paper addresses multi-class ransomware family classification by repurposing the Structured State Space Sequence (S4) model—designed for natural sequences—on static PE metadata. The authors reshape 71 tabular features into pseudo-sequences of length 32 and jointly train for binary detection and family classification, claiming S4 captures long-range dependencies in malware execution flow. They benchmark against CNN, LSTM, GRU, Transformer, and MLP baselines on the Ransomware Dataset 2024, reporting strong aggregate F1-scores. However, the study is limited to a single public dataset and single-run metrics, so its broader validity for IoT security remains unproven.

Spatial climate downscaling aims to recover fine-grained meteorological fields from coarse-resolution inputs, but standard pixel-level metrics such as MAE can hide critical failures in variability or spatial structure. This paper benchmarks five downscaling methods—from a residual U-Net to optimal-transport and contrastive-retrieval approaches—using five complementary criteria that disentangle pointwise accuracy, spectral fidelity, spatial alignment, distribution match, and gradient sharpness. The central finding is that no single method dominates across all measures, and method rankings systematically reverse depending on whether the evaluator rewards smooth alignment or realistic fine-scale variability. This motivates the core message: robust climate downscaling assessment must be explicitly multi-metric and application-aware.

LLMLeak demonstrates a covert channel where malware on a network-restricted device embeds secrets into URLs within fake error messages. When users paste these traces into LLM coding assistants, the model's legitimate web-fetching tool exfiltrates the secret via DNS or HTTP requests to attacker-controlled servers. The work challenges the assumption that local LLM deployment or prompt-injection defenses alone prevent data leakage to third parties.

Multimodal search agents traditionally interleave observations, freeform reasoning, and actions, but small models often waste inference budget generating lengthy reasoning traces that provide little actionable guidance. This paper introduces Selection-based Structured Reasoning (SSR), which replaces open-ended reasoning generation with a discrete selection among a small set of reusable natural-language reasoning candidates. By scoring all candidates in parallel via teacher-forced prefill with a shared KV cache, SSR cuts per-turn reasoning latency by over 90% while maintaining competitive task performance across seven benchmarks with 2B and 4B models.

Open-boundary hydrologic graphs such as river networks and coastal meshes exchange fluxes with unobserved exteriors, making long-horizon forecasting unstable because boundary forcing is unavailable at prediction time. This work proposes PRSOBG, which learns ghost-node proxies to approximate missing external forcing and applies local and global physics refiners to suppress error amplification during autoregressive rollout. Evaluated on the Danube river system and Chesapeake Bay, the framework cuts RMSE by up to 51% over a vanilla STGNN backbone with only about 8% training-time overhead.

Modern adaptive model development loops propose challengers against a moving incumbent: once a model is adopted, it becomes the reference for future candidates, meaning an unsupported replacement propagates into later hypotheses.

This paper introduces Evidence-Gated Research (EGR), a statistical adoption layer that freezes each challenger before its decision evidence is observed, constructs anytime-valid e-processes across predeclared environments under predictable routing, composes a persistent candidate e-value via an intersection–union minimum, and passes it to an online FDR controller.

The framework matters because it formally separates flexible proposal generation from the validity of state-changing adoptions and empirically demonstrates that naive development-only or single-environment validation can incur massive persistent false-discovery rates in closed-loop search.