Search AI news

Finite-Time Convergence and Regret Guarantees for Neural Reinforcement Learning in POMDPs

arXiv:2611.07069v1 Announce Type: new Abstract: Reinforcement learning (RL) in Partially Observable Markov Decision Processes (POMDPs) is computationally challenging due to the curse of long-range dependence on history. Fortunately, recent studies have shown that a range of POMDPs that satisfy the filter stability assumption can be well approximated by a superstate Markov Decision Process (MDP) based on a truncated history in a finite-memory window, with a value gap that decays with the window length. However, existing finite-time convergence analyses on such superstate MDP surrogates are either limited to over-simplified linear function approximation or rely on an i.i.d. sampling oracle that is violated by POMDP trajectories, both of which severely limit the applicability of these theoretical results in practice. This motivates us to bridge the gap by investigating RL in POMDPs with nonlinear neural value function approximation. Our main contributions are threefold: i) We first establish a new finite-state MDP point-wise approximation error bound, which lays the foundation for our RL algorithm design and analysis in POMDPs; ii) By applying Temporal-Difference (TD) learning to POMDP-generated superstate transitions and targeting the Bellman fixed point of the superstate MDP, we prove a finite-time critic error bound with nonlinear neural value function approximation; and iii) By combining the neural critic function approximation error bound with the policy mirror descent framework, we obtain a finite-time sublinear regret against the optimal POMDP value. To the best of our knowledge, such a finite-time sublinear regret based on neural function approximation is the first in the literature for finite-memory POMDPs. Our numerical experiments also verify our theoretical results.

Energy-Efficient Reinforcement Learning for Large Language Models

arXiv:2611.05748v1 Announce Type: new Abstract: Reinforcement learning for language models alternates between generating responses and training the policy. These stages place different demands on GPUs, creating opportunities to reduce clock frequency with little loss in throughput. Selecting frequencies online is difficult because response lengths change and frequency trials themselves consume time and energy. We present EE-RL (energy-efficient RL), an online controller that selects separate frequencies for generation and training during RL execution. It compares candidate settings with a prediction of the default GPU governor on the same token workload, and keeps the most energy-efficient setting that meets a throughput threshold. The controller then monitors throughput and energy and adjusts frequencies as the workload changes. Fixed-frequency sweeps on H100 and GB200 show GPU energy savings of up to 28.2% per valid training token across synchronous and asynchronous execution, while largely preserving throughput. EE-RL turns this headroom into online savings, reducing GPU energy per valid training token by up to 16.9% at a modest throughput cost.

Lesion Overlap Ranks PET Segmentation Models but Does Not Certify Patient-Level Tumor Burden

arXiv:2611.05564v1 Announce Type: new Abstract: Metabolic tumor volume (MTV) summarizes whole-body PET disease burden for prognostic stratification and treatment assessment. Automated MTV is useful only if its estimate is reliable for the individual patient, yet segmentation models are usually evaluated with overlap metrics such as Dice rather than burden error. We examined whether overlap metrics identify unreliable MTV estimates using 1,014 autoPET FDG PET/CT studies. We evaluated LesionTracer predictions held out from fine-tuning, compared five selected released models on 489 tumor-positive patients, and examined the findings in 597 PSMA studies. Across the five FDG models, ranking by higher mean Dice gave the same order as ranking by lower mean relative MTV error ($\rho=-1.00$, exact $p=0.017$). At the patient level, however, out-of-fold Dice tracked relative MTV error strongly ($\rho=-0.75$), yet 82 of 489 patients (16.8\%) fell on opposite sides of the Dice 0.80 and MTV-error 25\% thresholds. Within the training-exposed released models, correlations ranged from $-0.47$ to $-0.74$ over narrower, high-Dice ranges. Out of fold, opposing false-positive and false-negative volumes cancelled about one third of spatial error in net MTV. In 236 of 489 patients, each of the five models missed at least one lesion, though not necessarily the same lesion. Thus, agreement in average model rankings did not establish the reliability of individual MTV estimates or reveal shared lesion misses. PET segmentation benchmarks should report patient-level MTV error and missed-lesion counts alongside Dice and other overlap metrics.

From Visual Matching to Textual Conditioning: Cross-Domain Few-Shot Segmentation via Reference Inversion

arXiv:2611.04084v1 Announce Type: new Abstract: Cross-Domain Few-Shot Segmentation (CD-FSS) aims to segment objects from unseen domains using only a few annotated references. Recent Segment Anything Model (SAM)-based methods have advanced CD-FSS, yet their reference-target visual matching is highly sensitive to target appearance variations. SAM3 introduces text-conditioned segmentation, enabling natural-language descriptions to guide segmentation across target images, but directly using the category name as the condition still yields unstable performance under severe domain shifts. We propose Reference Inversion for SAM3 (RI-SAM3) that leverages SAM3's large-scale vision-language pre-training to learn suitable textual tokens for each target category. RI-SAM3 comprises three core designs: (i) Bi-level Reference Inversion that learns complementary tokens before and after the frozen text encoder: the former remain anchored in SAM3's text space while the latter represent the category more flexibly; (ii) Layer-adaptive Textual Conditioning combines these tokens differently as object queries evolve across decoder layers; and (iii) Presence-aware Confidence Supervision that jointly supervises query confidence and target presence for reliable predictions. With SAM3 frozen, RI-SAM3 optimizes around 12K parameters only per category and it efficiently segments each target in a single forward pass with reusable textual conditioning without per-sample reference-target visual matching. Extensive experiments over multiple widely adopted benchmarks show that RI-SAM3 achieves superior performance consistently in CD-FSS tasks, together with strong robustness, efficiency, and generalization.

Predictive gains, causal losses: why interventional competence cannot be learned from data alone

arXiv:2611.03861v1 Announce Type: new Abstract: Retaining predictive gains can require persistent errors in a model's response to interventions. For deterministic predictors with finite second moments under squared-error loss, any observational advantage over the best predictor using only an outcome's direct causes requires dependence on other inputs, regardless of the learning rule. When admissible interventions vary inputs while preserving the worldly outcome, this dependence yields a positive lower bound on root-mean-square intervention-response error. We identify conditions under which population gradient updates introduce this dependence from a parent-only baseline or damage a previously correct mechanism through shared parameters while improving other predictions. These dependencies can also make model behavior vary with how the same intervention is realized. We characterize interventional alignment through a common output mapping across internal states and physical processes realizing the same intervention target and relevant context. Correction is limited by the causal hypotheses currently admitted within inquiry. The relevant hypothesis class comprises these causal alternatives, rather than the function class the computational system could in principle represent. For predictions restricted to these alternatives, interventional mean error cannot fall below the best approximation they permit, whether evidence is observational or interventional. Reducing error below a positive limit requires admitting better hypotheses, which may entail revising the causal variables, mechanism structure, or intervention interpretation. Interventional criticism constrains replacements without determining them. We conjecture that reliably acquiring and maintaining interventional competence when causal commitments fail requires autonomous participation in proposing alternatives, testing their consequences, and revising them in response to error.

Rank-Aware Differentially Private Release of Listwise Preferences for LLM Alignment

arXiv:2610.37682v1 Announce Type: new Abstract: Listwise preference feedback offers richer supervision for large language model (LLMs) alignment than pairwise comparisons, but a full ranking also reveals more sensitive user preferences. Existing privacy-preserving alignment methods focus mainly on pairwise feedback, while listwise alignment under local differential privacy (LDP) remains unexplored. Extending LDP to this listwise domain raises a critical challenge: the high sensitivity of the full ranking space demands excessive noise, neutralizing the benefits of listwise supervision. To address this challenge, we analyze the interaction between supervision granularity and privacy noise in downstream training, and introduce a rank-aware exponential mechanism that privatizes listwise prefer

Artifact Distillation: Training-Free Transfer from Frontier to Small Language Models

arXiv:2610.37680v1 Announce Type: new Abstract: Closing the gap between frontier and smaller language models is usually approached by distillation: fine-tuning the small model (student) on a frontier model's (teacher) outputs, which binds transfer to training compute and leaves the transferred knowledge largely opaque. We propose Artifact Distillation: a teacher solves training instances and externalizes its problem-solving procedure into an artifact bank that a student queries at inference time, with no gradient updates. Across four LiveBench tasks, a bank written by Claude Opus 4.5 lets Claude Haiku 4.5 outperform unaided Claude Sonnet 4.5 on spatial reasoning (0.707 vs. 0.647) at 60% of its cost, and reach Sonnet 4.5's level on table joining at 89% of its cost. The same raw bank raises

Locating Sycophancy Across Training Stages in Open Language Models

arXiv:2610.37788v1 Announce Type: new Abstract: Language model assistants sometimes change a correct answer when the user suggests a wrong one. The behavior, called sycophancy, is usually attributed to preference-based post-training, yet early measurements found it in pretrained models. We ask where it comes from by measuring it at every available stage of training: 13 open checkpoints from 5 lineages, from base model through supervised fine-tuning and preference optimization to release. We separate how far an injected claim moves a model's answer from how firmly the answer was held, and vary who makes the claim, how reliable they are said to be, and how strongly they make it. Two sycophancy behaviors with different origins emerge, and both replicate across datasets. Deference to a claim i

ProtoEEG: Structured Prototype Alignment for EEG Visual Decoding

arXiv:2610.33065v1 Announce Type: new Abstract: EEG visual decoding remains difficult because scalp recordings are noisy, spatially coarse, and variable across trials and subjects. Most systems align each trial with a dense image embedding, although that embedding mixes semantic content with instance-specific texture, pose, and background. This target can demand details absent from EEG and provides no explicit account of the visual factors supporting a prediction. We propose \textbf{ProtoEEG}, a prototype-mediated framework that reformulates EEG--image alignment as soft composition over a shared learnable visual codebook. A trainable EEG encoder and a frozen visual teacher produce assignment distributions over the same $K=64$ prototypes. Their weighted reconstructions form a structured rep

Whispering Wisely: Smart Subset Selection for ASR Fine-Tuning

arXiv:2610.32997v1 Announce Type: new Abstract: Automatic Speech Recognition (ASR) models such as OpenAIs Whisper have substantially improved speech recognition quality but remain expensive to fine-tune, especially when large amounts of transcribed audio are available. This makes it challenging for many practitioners to adapt these models to new languages or domains. In this work, we introduce an embedding-based subset selection approach for selecting smaller training sets for ASR fine-tuning. Using Whispers pretrained encoder, we map each audio sample into a fixed-dimensional embedding space and apply a greedy k-center algorithm to select diverse and representative subsets of the training data. We evaluate this approach on both Hebrew and Danish, showing that k-center selection often impr

Import AI 478: Math for alignment; ProgramBench; and Mirendil shows what RSI means

What is “night” to a machine?

Distributional Alignment of LLMs and Its Generalization Properties

arXiv:2610.31687v1 Announce Type: new Abstract: Large language models are increasingly used to simulate and probe human behavior, e.g., for surveys. For these applications, a key property is that outputs satisfy distributional alignment: to match its output distribution with some reference (in this case, human) distribution. An obstacle in achieving distributional alignment is generalizing from small training data. In this work, we study (1) how to distributionally align LLMs; and (2) how aligned LLMs might generalize to unseen samples and unrelated topics and tasks. Using two large-scale benchmarks, we find that LLMs can indeed achieve distributional alignment and generalize to unseen questions of the same domain. Cross-topic generalization depends on the relationship between topics (e.g.

LazyFlash: Forward-Dense, Backward-Sparse Attention for Accelerated Long-Context Fine-Tuning

arXiv:2610.31911v1 Announce Type: new Abstract: As large language models (LLMs) process increasingly long sequences, attention becomes a major computational bottleneck. Sparse attention mostly targets inference, yet many methods still train densely or cannot support sparse training, and their added complexity makes sparsity hard to turn into wall-clock speedups, leaving dense attention widely used in practice. This paper introduces LazyFlash, a forward-dense, backward-sparse attention method built on the observation that attention training is dominated by the backward pass. LazyFlash keeps the forward pass exactly dense, so the trained model remains a standard dense model that existing systems such as vLLM can serve without any change. For the backward pass, LazyFlash proposes a gradient-a

Sebastian Raschka has released an eight-video course titled “Building a Reasoning Model from Scratch.”

On October 24, 2026, Sebastian Raschka published an eight-part, over-11-hour hands-on video course on Python and PyTorch in Ahead of AI magazine (magazine.sebastianraschka.com), systematically covering the evaluation of inference models, scaling during inference, reinforcement learning (RLVR), and knowledge distillation, with all methods implemented using the pre-trained Qwen3-0.6B model.

Why it matters：This course is the author’s advanced technical educational offering, following “Building Large Language Models from Scratch.” It focuses on the core evolutionary direction of today’s mainstream LLMs—enhancing inference capabilities—covering the entire technical pipeline from setting up the foundational environment to GRPO reinforcement learning and self‑distillation. The author explicitly states that all implementations are “hand‑crafted from scratch,” fully reproducible, and highly instructive with significant engineering reference value.

Amazon SageMaker AI has launched a Multi-Step Reinforcement Learning (MTRL) framework, supporting end-to-end training of agents to accomplish multi-step tasks.

AWS has launched the Multi-Turn Reinforcement Learning (MTRL) feature, enabling users to train large‑model agents on SageMaker AI using full task trajectories and final outcome feedback, without requiring step‑level annotations. A end-to-end demonstration is provided using Wordle as an example, covering AgentCore deployment, data preparation, GRPO training, reward design, and cost analysis.

OFA: Towards Building a One-For-All Model

2022 is a year of generalist models! With the bloom of multimodal pretraining, especially the unified model, we have witnessed the opportunity to building a generalist model that is capable of processing tasks of different modalities or multi-modalities! Thus, we propose OFA1, namely One-For-All, a unified multimodal pretrained model that unifies understanding and generation tasks concerning modalities into a single framework, and we pretrain OFA with the instruction-based multitask-pretraining that endows it with multiple capabilities.