Science & Mathematics

Learning motion planning policies that remain reliable under closed-loop execution is a central challenge in Autonomous Driving (AD). Imitation Learning (IL) scales well to large AD datasets but suffers from distribution shift, causal confusion, and compounding errors under closed-loop execution. In contrast, Reinforcement Learning (RL) typically requires handcrafted rule-based rewards and massive simulation. We study a simple closed-loop IL framework that uses RL purely as an optimization tool, guided by a trajectory-distance reward derived from logged expert data. We refer to this training strategy as Expert-Guided RL (EGRL). Through a controlled empirical study we find evidence that the training procedure, rather than model size and complexity, accounts for a substantial part of the performance gains. We then introduce DriveMLP, a lightweight MLP policy trained with EGRL, as a simple instantiation of these findings. It achieves a score of 86.51 on the nuPlan Val14 Reactive benchmark, a +3.2 point improvement over the previous best IL planner (Flow Planner), with >15x faster inference. Code is available at: https://github.com/alessandro-canevaro/DriveMLP .

Recursive self-improvement (RSI) describes a system that improves not only its behavior but also the process by which it improves. Recent digital agents test this idea by rewriting their own code and checking each version with executable tests. Bringing these mechanisms into the physical world requires learning from partial observations through costly trials, while accounting for changing operating conditions and potentially irreversible consequences. We position RSI within robot learning and trace pathways from improving robot capabilities to improving the processes that generate them. We examine how physical experience can strengthen these improvement mechanisms and when the resulting gains can support further self-improvement of the mechanisms themselves. Gains carry forward only when feedback addresses remaining learning needs, the receiving component can use the change, its value survives changes in tasks or hardware, and continued practice remains safe and affordable. Existing studies, concentrated in manipulation, support behavioral improvement, benefits to later learning, and procedure improvement. Some systems also change the improver's knowledge or tools, but the physical studies reviewed here have yet to isolate gains attributable to recursion. We synthesize these findings into evaluation principles and research directions for machines that, trial by trial, safely improve the way they improve.

Principal Component Analysis (PCA) is a fundamental tool for statistical dimension reduction, yet no canonical extension to distribution-valued data has emerged within the framework of optimal transport. While Wasserstein barycenters and Wasserstein least-squares regression provide canonical liftings of Euclidean averaging and linear regression in the Wasserstein space, an analogous principle for PCA has remained unexplored. In this paper, we introduce Lifted PCA, a canonical extension of Euclidean PCA to Wasserstein space, obtained through the framework of lifting functionals in the space of probability measures by convex extension. Starting from the classical variational formulations of PCA on the Stiefel and Grassmann manifolds, we derive two equivalent minimization problems over probability measures supported on these manifolds that correspond to Wasserstein barycenters with generalized costs. We establish an exact analogue of the equivalence between the Stiefel and Grassmann formulations of Euclidean PCA, preserving PCA's original variational structure under lifting. We further derive dual formulations and optimality conditions, and establish equivalences with multimarginal optimal transport and matching-for-teams problems. We also establish statistical consistency for empirical estimators and develop computational algorithms based on the Wasserstein barycenter formulation over the Stiefel manifold. Whereas Euclidean PCA summarizes vectors through projections onto an optimal low-dimensional linear subspace, Lifted PCA summarizes probability distributions through an optimal probability measure supported on multiple low-dimensional subspaces. Numerical experiments on distributional datasets illustrate the practical performance of the proposed methodology and its benefits for distributional data analysis.

Using an LLM to generate training data for another is among the most general ways to train one, and it remains common practice. Some frontier teams have closed that route for their own models: what comes back is an answer, or a short summary, without the detailed reasoning trace behind it. That is a little ironic: they were themselves built on distilling human knowledge, and distilling an open-source model is still entirely routine; the restriction applies to their own outputs alone. Recent work recovers the hidden reasoning by replaying encrypted blocks across models. It somehow works, but it breaks the providers' terms of service, so it is not something a product can be built on long-term. We built the alternative. We take the context and the teacher's answer as input, and generate the reasoning trace ourselves, in the style of the student's own model family. Training is two-stage, supervised and then adversarial: a generator learns to regenerate the family's own reasoning traces from their context and answer, and a critic learns how real a reasoning trace looks. Applied to a teacher's answer, the generator writes one in the family's style. An independent judge rates how likely traces are to be genuine on a scale from 0 to 1: ours average 0.68, real traces 0.66, and traces selected by the critic from four candidates 0.82. These are authenticity scores, not correctness rates. We test the idea on five benchmarks with everything else held fixed. On general tasks the five training corpora are about the same; on coding, the generated reasoning beats Fable 5's own thinking content. With the same 5,000 questions and the same teacher answers, HumanEval+ improves by 3.7 points and MBPP+ by 7.1 at the final checkpoint, and the MBPP+ lead holds at every checkpoint. We suspect that reasoning written in the student's family style is easier for it to learn from.

Pixel language models process text as rendered images instead of a finite vocabulary, which gives them strong cross-lingual transfer abilities. However, not much is known about how these abilities are influenced by larger-scale pretraining. In this paper, we train a family of MPIXEL models along three axes of variation: more languages, more scripts, and larger model capacity under fixed per-language compute. In evaluations on five tasks covering 83 languages, we find that scaling acts locally: each axis improves the added languages and scripts, while increased capacity mitigates the curse of multilinguality from pretraining on more languages. Transfer to unseen scripts improves with multilingual pretraining, but does not further improve with scaling. On typologically distant low-resource languages, our models outperform subword and character-level baselines across many writing systems. We also show that larger pixel models reserve more layers for visual processing, and that cross-lingual alignment weakens as more languages are added. Overall, our findings characterize how multilingual scaling improves pixel language models.

Modern vector search systems keep payload vectors in cloud object storage and only a compact routing index on fast local storage, typically using cluster-based ANN indexes. This design inherits a problem existing systems do not address. On object storage, small Range-GETs have latency dominated by fixed per-request overhead and are billed per request. Existing readers either issue one request per probed cluster, or coalesce adjacent ranges with hand-tuned gap or span thresholds that do not transfer across datasets or query workloads. We formulate the grouping of probed-cluster byte intervals into object-store requests as a global per-query optimization problem and present WaRP, a workload-aware request planner that solves it efficiently. Given the sorted byte intervals, a limit on the number of requests per query, and an estimate of how many object-store requests from one query can proceed concurrently under the target serving configuration, WaRP searches for a merge plan that minimizes a concurrency-aware latency surrogate while satisfying the request limit. Using an optimized dynamic program and adaptive frontier search, WaRP returns a plan within $(1{+}ε)$ of the optimum, with per-query planning overhead under $10$ms even in the most demanding tested settings. A lightweight offline reordering pass further improves request mergeability without changing the semantics of ANN search. We implement WaRP inside Lance's I/O scheduler and evaluate it on AWS S3. We build IVFPQ indexes on seven datasets, including SIFT100M and DEEP100M at the 100M-vector scale. WaRP + Reorder improves QPS by $1.09--3.48\times$ and reduces GETs by $2.12--5.19\times$ relative to Lance, while preserving Lance's recall.

Generative image compression (GIC) delivers strong perceptual quality at low bitrates. However, GIC methods with strong perceptual quality often rely on pretrained diffusion models with high latency, while recent lightweight GIC methods reduce latency but deliver lower perceptual quality. We observe that, for lightweight GIC methods at low bitrates, coarse deterministic quantization can collapse nearby encoder latents corresponding to different fine details to the same point, making these details difficult to recover. Motivated by this, we propose GR-GIC, an efficient GIC that delivers competitive reconstruction performance with fast inference. For better performance, we introduce Guided Refinement (GR), which combines latent refinement guided by entropy context with adaptive perturbations for training regularization, jointly improving reconstruction at low bitrates. For fast inference, we introduce reorganized projected residual vector quantization (P-RVQ), which equivalently reorganizes residual updates into low-dimensional lookups to expose greater parallelism and reduce sequential computation. Experiments on Kodak show that for performance, compared with OneDC, GR-GIC achieves BD-rate savings of 23.19%, 37.71%, 22.30%, and 22.14% for LPIPS, DISTS, PSNR, and FID, respectively. For fast inference, the highly optimized implementation of GR-GIC encodes and decodes in 4.60 and 5.98 ms on an NVIDIA A100. Specifically, the reorganized P-RVQ implementation achieves a $3.14\times$ average speedup over the original implementation under the same FP32 setting. On an NVIDIA Jetson Orin NX, GR-GIC achieves latency comparable to lightweight GIC methods.

Mixture-of-Experts (MoE) language models scale model capacity efficiently by activating a small subset of parameters per token. However, in long-context inference, memory remains a major bottleneck because the KV cache grows linearly with sequence length. Existing KV cache compression methods reduce this cost for dense transformers, but overlook a problem specific to MoE: KV eviction can change which experts are selected during answer generation, degrading the generation quality. We propose MoEZip, a routing-aware KV cache compression for sparse MoE language models. To estimate the sensitivity of expert routing to perturbation, we propose a novel metric based on the Fisher information matrix of routing distributions. MoEZip combines this metric with attention weights for routing-aware scoring of contexts. Finally, we compose a complementary scoring system that combines eviction methods for dense and sparse-MoE architectures. Experiments show that MoEZip achieves robust performance under aggressive cache compression on various benchmarks. At a KV retention ratio of 10\%, MoEZip improves over the SoTA baseline by +24 and +20 points on average on LongBench and RULER, corresponding to 2.3$\times$ and 4.9$\times$ higher scores, respectively.

The same classical f-divergence admits inequivalent extensions to quantum states. The question is whether these extensions can be obtained from a common classical variational principle with an operational interpretation. Here we generalize Matsumoto's reverse-test optimization by introducing retrodictive tests relative to a reference state. A retrodictive test compares a forward and a reverse stochastic process, realized by quantum preparation and measurement, related by Petz transposition relative to the chosen reference, and subject to prescribed quantum averages. Equivalently, it compares two reverse tests sharing a reference output, whose labels are coupled by retrodiction. For faithful states in finite dimension, we prove that the same spectral implementation attains the minimum classical f-divergence over all admissible implementations simultaneously for every operator-convex f. A flat reference yields the standard quantum f-divergences, thereby giving the Nussbaum-Szkoła distributions a variational characterization. Choosing either prescribed state as reference recovers maximal quantum f-divergences and Matsumoto's reverse-test optimization, while power references interpolate between the standard and maximal divergences along the geodesic family. In the relative-entropy case, this gives the passage from Umegaki to Belavkin-Staszewski relative entropy by changing the reference state. The same optimization extends to arbitrary quantum channels and, for channels with faithful Choi operator, gives a variational characterization of quantum entropy production.

Backpropagation remains the dominant training paradigm, using backward gradients and local updates that can suffer from vanishing or exploding gradients in deep networks. Although the Forward-Forward algorithm avoids conventional backpropagation, it remains mainly local or layer-wise, motivating global whole-network learning using forward evaluations only. For shared-kernel wavelet neural networks, we therefore propose a CMA-ES-based global forward-only learning framework. We further introduce structure-aware parameter decomposition and channel-specific credit assignment, factorizing the search into three subspaces. On BSDS500, our method reduces test MSE from 4.1783 to 4.0130 and improves PSNR from 42.621 to 42.787~dB over Adam/BP without backward propagation. Beyond these gains, training requires only 6 training-set coverages, compared with 1500 epochs for BP. Ablations show that structural decomposition and aligned credit assignment are more important than increasing population size, demonstrating the feasibility of whole-network forward-only learning for compact reconstruction networks. These results suggest a potential route toward forward-only optimization of compact reconstruction models for resource-constrained imaging and edge or near-sensor systems.

Large-scale aerial resources provide a foundation for learning navigation across diverse environments. However, existing resources remain limited in providing (1) densely sampled flight demonstrations for learning different policy formulations and (2) broad environment coverage for standardized closed-loop evaluation. To address these limitations, we introduce AirPlayground, a unified benchmark for training and evaluating aerial navigation policies across diverse environments. We first construct 94 photorealistic 3D Gaussian Splatting (3DGS) environments spanning indoor, urban outdoor, and vegetation settings, with aligned collision geometry. Building on these environments, we develop a scene-adaptive pipeline for automated expert flight collection, yielding 94K demonstrations and 46.5M synchronized observation-state-action samples at 50 Hz. We further provide an extensible API that supports training different policy formulations on this shared corpus and evaluating them under unified protocols in the corresponding environments. We benchmark seven baselines spanning four policy paradigms across three navigation settings in seen and unseen environments, and further examine scaling behavior with ACT. Scaling ACT's training corpus from 5 to 70 scenes raises median success rates on seen and unseen environments from 30.6% and 16.8% to 41.2% and 29.4%, respectively. Through this shared training and evaluation framework, AirPlayground supports analysis of how offline prediction errors relate to closed-loop outcomes. Our analyses highlight the roles of prediction errors, execution protocols, and scene interactions in closed-loop performance.

Pinching-antenna systems (PASS) flexibly reconfigure the wireless channel by positioning dielectric antennas along a waveguide near each user. Releasing and re-attaching an antenna, however, is a mechanical operation orders of magnitude slower than electronic switching, wasting much transmission time. Existing designs generally assume instant reconfiguration, neglect the mechanical latency, or keep the antenna fixed. This paper proposes S-PASS, a star architecture of waveguides sharing one radio-frequency chain through a switch, and pipelined antenna pre-positioning (PAP), a time-division protocol in which idle waveguides pre-position their antennas for their next users while one transmits. It derives in closed form the scheduling efficiency, the number of waveguides beyond which the latency costs no transmission time, and the latency up to which PAP outperforms fixed placement. Analytical and numerical results show that this latency grows with each added waveguide, and users at the outer ends of the waveguides gain most.

Policy engines determine whether evaluated policies permit a request, but a correct ALLOW decision does not establish complete governance-source participation, exact context binding, or sufficient grounds for execution admission. This paper examines the gap between policy evaluation and execution admission. We introduce the Binding-Complete Execution Admission Contract (BCEAC), a governance-relative contract for independently determined source obligations, exact policy and contextual binding, and typed admission outcomes under explicit trust assumptions. We propose a conditionally engine-agnostic architecture in which participating backends must satisfy declared semantic profiles, adapter contracts, and contextual correspondence obligations; an Execution Boundary Language (EBL) authoring syntax remains optional. The architecture separates governance resolution, policy evaluation, independent admission verification, and downstream execution mediation. We investigate two distinct experimental tracks: a finite Governance Source Completeness study and an OPA/Cedar experiment with archived native-engine records examined by offline review. The OPA/Cedar archive comprises 72 scenarios and 144 strict case-engine records (120 native-invoked and 24 pre-native), with eight additional permissive controls outside the strict denominator. The offline review did not rerun the original native OPA/Cedar executables. The results provide bounded evidence while retaining historical failures and unresolved implementation defects. The work does not establish constitutional adequacy, general semantic refinement, authenticated currentness, machine-checked execution-safety theorems, or protected external effects. Its contribution is an explicit, falsifiable framework for determining what must be established between a policy decision and an execution-admission judgment.

We present SpatialDex, a framework that learns 3D-grounded bimanual dexterity from RGB-only human videos through reconstruction-augmented visuomotor pre-training. The central idea is to learn human behavior together with its spatial context. From each video, we reconstruct the scene geometry and estimate the hand motion, align them at a shared video-level scale using the hand-size prior of 3D hand estimates. We retain the reconstructed geometry as both a policy input and a prediction target. The policy grounds the token grid of a pre-trained vision-language model in this geometry through gated residual fusion, is trained to predict future scene geometry alongside wrist and finger actions, and generates finger motion conditioned on the predicted trajectories of both wrists. Pre-trained on 1M human-video clips and fine-tuned with a small number of robot demonstrations, SpatialDex outperforms the state-of-the-art methods by 16.5 points on nine simulated dexterous tasks across two embodiments and by 38.3 points on six real-world bimanual dexterous tasks. Under a matched comparison, pre-training with reconstructed geometry improves success by 14.2 points over pre-training on hand motion alone.

Spiking Neural Networks (SNNs) promise energy-efficient inference through sparse, event-driven computation, yet remain constrained on continuous, context-dependent tasks such as automatic speech recognition (ASR). A key bottleneck is the Leaky Integrate-and-Fire (LIF) neuron, whose membrane decay erases information across pauses and phonetic boundaries while it must both extract features and retain long-range context. Inspired by cortical separation of nonlinear filtering and somatic integration, we introduce the Spiking Ternary GRU (stGRU): a ternary-weight gated recurrent unit performs temporal filtering, while downstream Resonate-and-Fire (RF) neurons generate the binary spike stream driving the next block's event pathway. This makes binary event transmission a native training constraint rather than a separate deployment-time activation quantization step; together with ternary weights, learned projections avoid dense floating-point matrix MACs. Trained end-to-end on LibriSpeech-960, Small, Medium, and Large stGRUs reach 8.09%, 7.04%, and $6.44 \pm 0.17\%$ WER on test-clean without an external language model. To our knowledge, these are the first matrix-MAC-free spiking networks without a learned ANN sequence decoder to achieve single-digit WER on LibriSpeech. The architecture also achieves 10.88% test CER on Aishell-1 (9.28% on dev) when trained from scratch. Internally, trained networks develop a depth-graded hierarchy: shallow layers act as fast acoustic filters, while deep layers form slower semantic integrators whose latent representations segregate into part-of-speech manifolds and word-level spike events. Deep-layer firing rates follow lognormal distributions with a sparsity gradient, mirroring hierarchical processing observed in human cortex in vivo. Our biologically grounded architecture combines competitive ASR with favorable modeled compute and memory cost.