Search arXiv⌕ Search

arXiv subjects

Jun Wang

Publications and source records attributed to Jun Wang.

At least 37 records · Page 2Linked to original sources
arXiv2026

Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks

Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives. Yet limited observations can support multiple world models that explain past interactions but predict different outcomes in unseen states. We introduce Memento 3, building on the Memento series to enable frozen LLM agents to continually learn explicit world models through external memory. The agent maintains a natural-language rulebook as persistent semantic memory, recording revisable hypotheses about environment dynamics while leaving unknown aspects underspecified. It compiles this rulebook into executable code for prediction and planning. Through a continual loop of observation, reflection, rule revision, compilation, and verification, the agent uses prediction errors to refine both the rulebook and its code. Updated code is accepted only when the LLM judges it faithful to the rulebook and cell-exact replay reproduces the observed transitions. We investigate this process as a model-based route to recursive self-improvement (RSI): the agent autonomously explores the environment, revises its world model, and uses verified updates to guide subsequent interaction and learning, while the underlying LLM remains fixed. A population extension maintains multiple world models in parallel, sharing interaction evidence and using their predictions to guide exploration. On ARC-AGI-3, the single-model agent clears every level of all 25 public games, achieves a mean Relative Human Action Efficiency (RHAE) of 100.0, and uses 44% of the human action count. In an Atari Pong case study, a learned feedback controller wins 21:0 in each of three evaluated episodes with different openings, without further LLM calls.

cs.AI↗
arXiv2026

Normalized solutions to critical and supercritical Schrödinger equations in a ball

In this paper, we study radial normalized solutions to nonlinear Schrödinger equations in the unit ball \[ -Δu+λu=|u|^{p-1}u, \qquad \int_{B_1}|u|^2\,dx=ρ, \] where $λ$ is unknown and $p\ge p_c:=\frac{N+2}{N-2}$. Soave \cite{Soave} proved that for every $1 p_c$. Thus, in the radial class, the subcritical all masses phenomenon fails throughout the full Sobolev critical and supercritical range.

math.AP↗
arXiv2026

A self-learning scientific agent for X-ray diffraction

A central challenge for scientific agents is to turn analytical experience into reusable expertise grounded in physical evidence. Here we introduce Gan Jiang, a self-learning agent for powder X-ray diffraction built on a diffraction-analysis ecosystem we developed: XMatcher, XQueryer, XDecomposer and WPEM. Together, these engines span phase identification, multiphase decomposition and physics-constrained whole-pattern modelling. Gan Jiang converts analytical experience into executable skills by diagnosing failures, revising skill instructions and code, and validating revisions before reuse, without retraining the language model or changing the underlying physical models. Skills selected using development data and frozen before held-out evaluation achieve higher refinement scores than the original expert-designed skills across FullProf, GSAS-II and PyWPEM. The agent resolves strongly overlapping reflections, quantifies a five-phase ancient Egyptian cosmetic, tracks lattice evolution in an operating battery and compares atomic configurations in a disordered oxide catalyst. On DeltaXRDbench, it leads the evaluated methods in single- and multiphase identification across simulated and experimental data. Without supplied composition, single-phase top-1 accuracies reach 96.30\%, 81.78\% and 40.83\% on MP500, RRUFF and opXRD, respectively, compared with 58.00\%, 58.47\% and 26.45\% for the strongest comparator. These results demonstrate how an integrated scientific tool ecosystem can support agents that extract structural knowledge from measurements while accumulating validated analytical expertise that transfers to new samples.

cond-mat.mtrl-sci↗
arXiv2026

Projection of purification performance for the RELICS experiment

The RELICS (REactor neutrino LIquid xenon Coherent elastic Scattering) experiment employs a dual-phase liquid xenon time projection chamber to search for Coherent Elastic Neutrino-Nucleus Scattering (CE$ν$NS) induced by reactor neutrinos. To detect these sub-keV nuclear recoils and minimize signal attenuation, it is critical to maintain a sufficiently low impurity concentration in the detector. This work presents a comprehensive purity evolution model developed to describe impurity migration inside the detector. Utilizing measured material outgassing rates as input parameters, the model incorporates non-uniform transport mechanisms of the impurities, including circulation, vaporization, and condensation. The model is validated using data from a dedicated prototype detector. Based on this validated model, projections for the purification performance of the upcoming RELICS-10 and RELICS-50 detectors are provided.

physics.ins-det↗
arXiv2026

Retrospective Progress-Aware Self-Refinement for LLM Agent Training

Long-horizon LLM-based agents receive rich environmental observations during interaction, yet outcome rewards provide limited explicit supervision about how individual actions advance task completion. We investigate whether agents can turn this interaction evidence into useful training signals through retrospective progress assessment. A WebShop pilot shows that direct progress prompting reduces task success, whereas hindsight-annotated demonstrations improve it. We introduce RePro, Retrospective Progress-Aware Training, with a forward-then-reflect rollout: the agent estimates progress while acting, then reassesses each step using the completed trajectory and outcome. After warmup with externally generated demonstrations, policy optimization combines self-generated progress differences, online-retrospective alignment, and format rewards with environment feedback, requiring neither a separate process reward model nor ongoing teacher annotation. Experiments on WebShop, ALFWorld, and Sokoban show that RePro enhances the Qwen family's performance, with up to 11.57% success rate gains.

cs.CL↗
arXiv2026

Language models can notice an impossible engineering problem yet still report it as solved

Language models draft engineering calculations, but answer accuracy does not show whether they reject an impossible problem. We tested 14 models on 30 pairs of mechanics problems, each with a valid version and one made impossible by changing a given value or assumption. Two independent solvers verified every answer key and showed that each flawed problem was physically impossible. We scored solving of valid problems separately from rejection of their flawed counterparts. Each reply required a "solved" or "cannot solve" status; rejection meant "cannot solve" or withholding an answer. The initial prompts did not warn that problems could be flawed. Across three recent models, 12 of 90 replies failed to reject a flawed problem. In 11 of these replies, the model stated the flaw, answered a corrected problem and still reported the original as "solved", according to artificial intelligence raters and numerical checks. We later retested four models from one provider, offering "flawed" instead of "cannot solve" and asking them to name and explain the defect. Three models showed statistically significant increases in rejection, but valid-problem solving fell in three. Evaluations therefore need to score both versions and distinguish flaw recognition from the reported status.

cs.AI↗
arXiv2026

AMBER: Multi-View Adaptive Budget Allocation for Listwise Vision-Language Reranking

Vision-language models (VLMs) are powerful listwise rerankers for multimodal retrieval, but high inference costs restrict them to evaluating small local candidate views. Existing multi-call strategies rely on fixed schedules, wasting expensive VLM calls on uninformative candidate pairs and easy queries. To address this, we propose Adaptive Multi-view Budgeted Elo Reranking (AMBER), an online, budgeted multi-view reranking framework that dynamically optimizes global resource allocation. AMBER treats fragmented listwise VLM outputs as local tournaments, using continuous Elo updates to maintain a lightweight global ranking state. Building on this, it allocates computation at two levels: dynamically constructing candidate views with high score ambiguity, and scheduling queries to maximize expected information gain. We show that each Elo update corresponds to a stochastic gradient ascent step on the Bradley-Terry log-likelihood, and provide a submodular information-theoretic motivation for the query-level allocation strategy. Experiments on CIRR, CIRCO, and PhotoBench demonstrate that AMBER achieves the strongest overall performance among the compared multi-call VLM reranking methods under comparable VLM-call budgets, while remaining effective in lower-budget settings. Our code is publicly available at https://github.com/wnlfc/AMBER.

cs.AI↗
arXiv2026

MagLearn 2: High-Fidelity, Saturation-Aware, and History-Efficient Sequence-to-sequence Modeling of Transient B-H Behaviour Under PWM Excitation

Accurate transient B-H modeling under pulsewidth-modulated (PWM) excitation is challenging since the magnetic-field response depends on both the instantaneous flux density and the preceding magnetisation trajectory. Although sequence-to-sequence models enable waveform prediction, existing studies have not jointly addressed how much magnetic history is required, how saturation region should be handled, and how trained models can be adapted to unseen materials. To address these gaps, this paper presents MagLearn 2, a condition-aware, short-window LSTM-based sequence-to-sequence framework for reconstructing the future H(t) response from the historical B(t) and H(t) trajectories. A saturation-aware auxiliary module selectively replaces saturation-dominated predictions within the complete waveform forecast. In a representative saturated 3C90 case, the auxiliary module reduces the sequence-level relative RMSE from 67.84% to 9.19%. To understand how much the history samples contributes to the predictions, an occlusion-based sensitivity analysis is conducted to quantifies the contribution of magnetic history and reveals that that the information most relevant to prediction is concentrated near the forecast boundary. In the independent evaluation for MagNet Challenge II, the proposed framework's best-performing model achieved the lowest reported average RMSE to date on the sequence-prediction task at 2.2%.

eess.SY↗
arXiv2026

Superficial Reflection or Genuine Thought? A Fine-Grained Cognitive Analysis of Large Reasoning Models

Motivated by the observed human-like behaviours in Large Reasoning Models (LRMs), this paper introduces a comprehensive taxonomy to characterise atomic reasoning steps and analyse the reasoning behaviours of LRMs. Grounded in human cognitive processes, we propose a taxonomy comprising five groups and seventeen categories. Through this taxonomy, we conduct an in-depth analysis of contemporary LRMs and distil four actionable takeaways for model optimisation. Most notably, we reveal that prevailing post-answer ``doublechecks'' are largely superficial and rarely yield substantive revisions. A targeted intervention further shows that explicitly eliciting richer reflection processes can substantially improve failed self-correction. To support this largescale study, we propose CAPO, an automated annotation method used to construct a dataset of 277,534 reasoning steps with strong agreement with human expert annotations. We further validate the main behavioural patterns on a newer reasoning model and a coding domain, demonstrating the broader applicability of the proposed taxonomy. All source code and data are available at https://github.com/hehepig4/psyche.

cs.AI↗
arXiv2026

Global dynamics near solitary waves for the generalized Boussinesq equation under even-odd perturbations

We study the global dynamics of the generalized Boussinesq equation near standing solitary waves. For even-odd perturbations, Maulen [J. Math. Pures Appl. 177 (2023)] constructed an asymptotically stable center-stable manifold M near the solitary wave Q. We complete this local description by proving that any sufficiently small even-odd perturbation of Q lying outside M either scatters in the energy space as t -> +infinity or blows up in finite time. The proof relies on two main ingredients. First, we classify even-odd solutions with energy below the ground-state energy by identifying two invariant regions, associated with scattering and blow-up, respectively. Second, we establish a one-pass theorem in one dimension: a non-scattering solution cannot re-enter a suitably chosen neighborhood of the solitary wave once it has left that neighborhood. Consequently, the solution remains confined to one of the two invariant regions. Together with the classification above, this yields a dichotomy between scattering and blow-up outside M.

math.AP↗
arXiv2026

Harnessing Large Language Models to Compile Task-Relevant Context into Bayesian Optimisation

Incorporating rich task-relevant context, such as domain knowledge and external observations, is a key capability yet remains challenging for Bayesian optimisation (BO). Recently, practitioners have started to use large language models (LLMs) to generate and execute BO programs through coding harnesses. In such emerging practices, the posterior belief is shaped not only by Bayesian inference but also by LLM-generated model and data artefacts, offering a flexible route for task context to enter BO as executable code. To study whether and how LLMs can be harnessed to compile diverse contextual signals for BO, we formulate LLM-compiled BO as generalised-context decision making. We propose HarBO, a BO-specialised harness that compiles generalised context into the core artefacts of standard BO through a validated multi-stage workflow. Our theory analyses the regret under imperfect compilation and the effect of adding new context. Across synthetic functions and real-world benchmarks, we find that LLM harnesses can effectively compile context into standard BO, achieving competitive performance with specialised LLM-embedding-based and direct LLM-in-the-loop BO methods. General coding harnesses can be effective in familiar domains such as hyperparameter optimisation, but fall short in unfamiliar, context-rich domains. Together, these results establish LLM harnesses as a promising, but not automatically reliable, route for making rich task context usable in BO.

cs.LG↗
arXiv2026

AVSG: Accelerated Vectorized Sparse Gather for Efficient KV Cache Offload in Sparse-Attention LLM Serving

Dynamic sparse attention reduces long-context attention computation by selecting only a subset of tokens, but still requires access to the full KV cache, leaving serving memory-bound. Offloading the KV cache to host memory reduces device memory pressure but places H2D transfers on the decoding critical path. In DSA, substantial overlap in selected KV entries across decoding steps creates an opportunity for device-resident reuse. Exploiting this reuse efficiently, however, presents three critical challenges: costly matching of selected tokens against entries retained in the HBM buffer, uneven distribution of the remaining H2D transfers across accelerator cores, and retaining frequently accessed KV entries within limited HBM capacity. We present AVSG, an operator that addresses these challenges for DSA while preserving its exact token selections. AVSG uses vectorized hash matching to identify reusable HBM-buffer slots and reserve slots for entries requiring transfer, then evenly partitions these H2D transfers across accelerator cores. Lifetime-based buffer management retains frequently accessed KV entries in the device buffer across decoding steps, and shared slot reservations extend reuse across tokens within a multi-token prediction iteration. On a single NPU, vectorized hash matching is 2.80x faster than scalar dual-pointer matching, miss-only transfer raises effective H2D bandwidth by up to 34.44x over request-level assignment, and an 8K-entry buffer reaches a 94.83% hit rate on real requests. These gains translate into end-to-end improvement: on a serving stack processing a real-world production dataset, AVSG reduces time per output token by 39% and increases output throughput by 1.27x relative to the same KV offload layout without HBM-buffer reuse, demonstrating the benefit of efficient matching, balanced transfer, and effective HBM residency in production-scale serving.

eess.SP↗
arXiv2026

iSDFT: Information-Proximal Self-Distillation for Continual Learning in LLMs

On-policy self-distillation fine-tuning (SDFT) learns new skills from demonstrations while reducing forgetting, but it always distils toward the full demonstration-conditioned teacher. This fixes teacher influence at the full-teacher endpoint, providing no control over how much demonstration information should be transferred at each prediction state. We introduce Information-Proximal SDFT (iSDFT), which instead treats the teacher as a budgeted source of information. At each token, iSDFT selects the distribution closest to the current student that satisfies a prescribed teacher-information constraint, yielding a closed-form exponential target with a locally determined tilt. To control cumulative drift, we further anchor the student to its frozen base policy. Across four heterogeneous LLM backbones and two specialisation tasks, iSDFT improves vanilla SDFT in 7 of 8 model-task settings and matches it in the remaining one. It also provides tighter retention on the original SDFT benchmark suite, with 73% of evaluations remaining within 0.5 points of the base model versus 52% for the strongest baseline, while achieving the largest mean improvement on all ten additional mathematics, coding, and competition-mathematics benchmarks. These results show that controlling how much and when teacher information is introduced improves specialisation while preserving broader capability.

cs.LG↗
arXiv2026

HiThink Turn: An Intent-Aware Turn-Taking Control Module for Full-Duplex Dialogue

Full-duplex dialogue requires timely yet selective interruption handling, which end-of-turn prediction alone cannot achieve: complete utterances may need no response, while unfinished requests may warrant interruption. To address this challenge, we propose HiThink Turn, an intent-aware streaming turn-state predictor that separates response intent from semantic completeness and conditions decisions on system playback state. A key contribution is minimal intent-sufficient prefix supervision, constructed through LLM judgments and speech alignment, while training on audio truncated at chunk boundaries improves robustness to partial speech. These components support streaming inference with 240-ms audio chunks, enabling low-latency, accurate full-duplex turn control. Experiments show that HiThink Turn leads the compared methods in Easy Turn macro accuracy, Full-Duplex-Bench average interaction rate score (0.933), and non-target-speech average playback resume rate (0.735). Additionally, intent-prefix triggering raises interruption success from 89\% to 98\% and reduces mean stop latency by 60.9\%.

cs.HC↗
arXiv2026

Identification of $D_{s1}(2933)$ as the $2P_{1}$ state in the quark model

We investigate the internal structure of the newly observed charm-strange meson $D_{s1}(2933)^{+}$, which was reported by the LHCb Collaboration in $B^{0}\to D^{+}D^{-}K^{+}π^{-}$, in the quark model by treating it as the $2P_{1}$ state of the $c\bar{s}$ system. The coupled-channel corrections to the ${}^{1}P_1$--${}^{3}P_1$ mass matrix are evaluated through the two-point self-energy loops involving the $DK^{*}$, $D^{*}K$, $D^{*}K^{*}$, $D_{s}ϕ$, $D_{s}^{*}ϕ$ and $D_{s}^{*}η$ intermediate channels. Requiring the corrected eigenvalue to reproduce the measured mass yields a cutoff scale of about $0.82~\mathrm{GeV}$, a $2P_{1}^{\prime}$ partner state at $3012~\mathrm{MeV}$, and a mixing angle of $-12.9^{\circ}$ that deviates significantly from the heavy-quark ideal-mixing angle. Moreover, the predicted three-body branching ratio fraction $R_{\text{th}}=1.98$ for $D_{s1}(2933)\to D^{+}K^{+}π^{-}$ via the $DK^{*}$ and $D_{2}^{*}K$ intermediate states is consistent with the experimental value $R_{\text{exp}}=1.27^{+1.29}_{-0.95}$ within uncertainties. These results support the assignment of $D_{s1}(2933)$ as the $2P_{1}$ excited state of the $c\bar{s}$ system with final state interactions.

hep-ph↗
arXiv2026

Reward-Decomposed Reinforcement Learning for Immersive Video Role-Playing

Text-based role-playing models can imitate character styles, but often fail to capture scene atmosphere and evolving tension, which are crucial for immersive applications such as VR games and interactive narratives. We study video-grounded role-playing dialogue and introduce EBM-RL (Eye--Brain--Mouth Reinforcement Learning), a decoupled GRPO-based framework that separates observation ( ), reasoning ( ), and utterance generation ( ). This design mimics the human See-Think-Speak process, enabling the model to ground dialogue in visual perception before reasoning and response generation. To optimize this See-Think-Speak process, EBM-RL integrates complementary rewards for scene--text alignment, perceptual--cognitive utility, answer faithfulness, and format consistency. Extensive experiments show that EBM-RL substantially outperforms text-only role-playing baselines and larger-scale vision-language models on our immersive role-playing benchmark, improving both visual-atmosphere consistency and character authenticity. Moreover, EBM-RL demonstrates strong zero-shot transfer to out-of-domain VideoQA benchmarks without additional fine-tuning. We also release an open-source dataset for video-grounded role-playing dialogue.

cs.AI↗
arXiv2026

MILO: Efficient Many-shot In-Context Learning with Block-wise Low-rank Compression

Many-shot in-context learning (ICL) enables large language models (LLMs) to adapt to complex tasks by conditioning on thousands of demonstration examples, but this paradigm shifts the inference efficiency bottleneck to the key-value (KV) cache memory. Due to the linear scaling behavior of the KV cache, storing these intermediate tensors has become a paramount challenge for both online serving and on-device deployment. To address this issue, we propose a novel compression framework, termed MILO, that exploits the low-rank redundancy inherent in many-shot contexts. Specifically, MILO features a block-wise low-rank compression strategy that compresses the KV cache at the block granularity, where each block contains multiple many-shot examples. Furthermore, to handle the heterogeneous context density across different blocks, MILO dynamically allocates rank budgets based on the information entropy, preserving the fidelity of critical blocks while aggressively compressing redundant ones. Experimental results on Qwen2.5 models demonstrate that our method achieves up to 50% reduction in KV cache memory and 1.8x throughput improvement, with negligible performance degradation on classification and reasoning benchmarks, significantly outperforming prior baselines.

cs.CL↗
arXiv2026

What Makes an Efficient VLA? Navigating Action-Head Design, Scaling, and Latency

Vision-Language-Action (VLA) models combine a pretrained vision encoder, a language backbone, and an action head, but their relative contribution has not been established under controlled, latency-paired conditions. We fix the backbone families (SigLIP2 and Qwen2.5) and the training pipeline, sweep action-head design and module scale, and pair each configuration with measured on-device latency. The study yields three findings. First, action-head performance is governed primarily by initialization rather than decoder architecture, loss, or inference budget: copying the last transformer layers of the language backbone into the head is the single largest lever, at no latency cost, and the only axis that helps at every module scale. Alignment also explains the other axes: flow matching and a heavier decoder pay off only while the head is misaligned and reverse once it is aligned, and extra inference passes give no measurable benefit; expressiveness appears to substitute for missing alignment. We read this as representation transfer: the aligned head keeps attending to the instruction's object nouns and stays close to the backbone in weight space rather than relearning to act from scratch. Because we reach alignment only through initialization, we offer this as the account that best organizes the measurements, not a demonstrated cause, and name the control that would settle it. Second, capacity pays only after alignment: the aligned action head is the highest-return module to scale. Third, those returns diminish sharply near the size today's $π$-series VLAs already use, so further growth buys little in-domain accuracy for its latency. These specify EffVLA, a compact model matching the strongest open-source VLAs on standard LIBERO, leading on most LIBERO-Plus perturbation axes at lower latency, and transferring to a real SO-ARM101 arm with the recipe unchanged.

cs.RO↗