#01Oct 19, 2026
cs.RO
Where is Physical RSI capability from? An Observational Scaling Law of LLM-as-policy
Zekai Li, Jiayi Yu, Jinlu Zhang and 8 more
In this paper, we systematically investigate the mechanisms, scaling behavior, and generalization of physical Recursive self-improvement. We introduce PhysRSI-eval to quantify physical RSI through a five-axis metric: First-Attempt Performance, Efficiency, Success, Budget and Improvement from failure. Different VLMs are evaluated on 18 manipulation tasks under a fixed interaction budget. The analysis of experimental results motivates the Physical Barrel Hypothesis: different capabilities may constrain different parts of improvement. The five-axis metrics also distinguish systems with close public benchmark averages and show that high scores need not come with early or efficient gains. We study observational scaling from two perspectives: predicting the relative performance of VLM agents from their public capabilities, and forecasting later progress from early physical feedback. Together, these analyses ask how quickly and how far an agent can improve. Moreover, we conduct workspace interventions and then analyze what improvement leaves behind. Removing history reduces late-round success, frozen workspaces remain useful on unseen layouts, and three cross-task cases achieve success earlier. The results suggest that scene-specific adaptation and reusable experience can coexist.
#02Oct 19, 2026
stat.ML
Teacher Supervision, Fixed Representations and Calibration: Exact Risk Calculations for Candidate-Scoring Decision Models
Agus Sudjianto, Denis Burakov
Candidate-scoring decision models return a probability distribution over a presented set of answers. Public documentation of one such system, Jev, and of an open counterpart, JevAny, describes three training ingredients that determine what these probabilities estimate: soft targets from larger models, a frozen pretrained backbone and temperature calibration. We analyze each ingredient in Gaussian surrogates where the finite-sample risk can be computed exactly. A student trained on teacher targets inherits the teacher's bias, which more student data does not remove and agreement with the teacher does not reveal. A fixed representation lowers the risk only below an explicit sample-size threshold. In a probit model the exact recalibration map is a temperature scaling, which removes all excess risk only when the teacher's coefficient has the correct direction and otherwise leaves a residual that calibration diagnostics do not detect. Under Luce's choice axiom, candidate scores are coherent across refinements of an event if and only if they aggregate by log-sum-exp. On 231,521 credit card complaints from the Consumer Financial Protection Bureau, where the decision is whether a complaint ends in monetary relief, a student trained on a logistic teacher's probabilities ends at the teacher's decision cost while a student trained on outcomes goes 8% below it, a 30B model read zero-shot remains worse than the teacher after recalibration, and that model departs from the coherence condition by 17.5 decibans. Pretrained representations reduce the labeled data required and calibration corrects the scale of the probabilities, but neither corrects a biased teacher, so the probabilities have to be checked against realized outcomes.
#03Oct 19, 2026
stat.ME
High-dimensional covariate-assisted ranking
Yujia Liu, Yuanyuan Lin, Ruijian Han and 1 more
Ranking data in modern applications often involves covariate vectors whose dimension grows with the number of objects. Identifying relevant covariates and integrating them with item-specific utilities can yield better predictive performance. In this work, we study a Plackett--Luce model that incorporates both item-specific utilities and high-dimensional covariate effects. Under a sparsity assumption on the covariate coefficients, we establish a necessary and sufficient condition for identifiability in terms of the Kruskal rank of the restricted covariate design. We then propose an $\ell_1$-regularized maximum likelihood estimator for the covariate coefficients. Under suitable design and connectivity conditions, we establish the existence and uniqueness of the estimator with probability tending to one, as well as its uniform consistency. With an additional minimum-signal condition, we further establish exact support recovery. Our analysis combines a hypergraph-based chaining technique with a profile-likelihood perturbation analysis, yielding estimation rates that explicitly capture the effects of the comparison hypergraph's connectivity and the underlying sparsity level. Numerical experiments using synthetic data and a horse racing dataset corroborate our theoretical findings.
#04Oct 19, 2026
cs.AI
AnyGoal: On World Model Generalization Beyond Expert Rollouts
Heejeong Nam, Divyat Mahajan, Quentin Le Lidec and 2 more
The goal of a world model is to capture the dynamics of an environment. This has the great appeal that it allows agents to plan and simulate a course of actions with the help of a world model. For these uses to be reliable, however, the learned dynamics must behave well beyond the specific state-action combinations observed in expert rollouts. Despite this, the relationship between data coverage, dynamics generalization, and downstream planning performance remains underexplored. We study this relationship through the lens of compositional generalization: can a world model recombine familiar state and action factors into valid dynamics? To answer this, we introduce AnyGoal, a benchmark of goal-shifted task variants built from widely used world-model environments, together with an open-source library. Each task is designed to be within reach of a world model that has learned the underlying dynamics: individual state factors and actions remain within their training support, but their combinations are underrepresented in expert trajectories. Across multiple environments and world model families, we find that current models consistently exhibit failures under these recombinations. By examining dynamics and representation failures across model families, interaction regimes, and evaluation metrics, we study where current world models generalize and where challenges remain for goal-shifted prediction and planning. Together, this framework, benchmark, and analysis provide a basis for evaluating progress toward world models that generalize to new planning tasks.
#05Oct 19, 2026
cs.LG
On Data Scaling in Tabular ML: More Data, Better Generalization?
Ivan Rubachev, Nikita Lazarev, Artem Babenko
Recent work in tabular machine learning has emphasized the importance of diverse and realistic evaluation settings. The niche of large-scale (1M+) data, however, remains under-explored. To study training-data scaling, we assemble eight real-world tabular datasets with up to around ten million training examples, retaining the natural distribution shifts of the tasks and using random-split evaluation as a control. We find that GBDTs and MLP-based models show broadly similar scaling trends, with gains from both additional training data and algorithmic improvements. Tabular foundation models show more varied scaling behavior, with recent versions benefiting more consistently from larger contexts even under distribution shift. However, scaling training data under distribution shift introduces new challenges: in-distribution and out-of-distribution performance mostly improve together at smaller scales but can diverge at larger scales. We examine this through MLP capacity sweeps and comparisons of the validation distributions used for tuning and early stopping in MLPs and GBDTs. We also compare common practices of using validation data, contrasting its use for model selection in MLP pipelines with using it for in-context fitting in tabular foundation model pipelines. Under stronger shifts, MLPs selected using validation representative of the evaluation conditions generally maintain or improve performance as training data grows. In contrast, including such examples in context improves foundation-model performance, but does not prevent deterioration with increasing training size in our experiments.