arXiv subjects

Kaijie Chen

Publications and source records attributed to Kaijie Chen.

At least 19 recordsLinked to original sources

arXiv2026

Convergence Theory of Knowledge Distillation in Asynchronous P2P Gossip Learning Network

Decentralized, serverless learning increasingly connects devices running different architectures, where the standard tool, decentralized SGD, is undefined as models with different parameter counts cannot be averaged. Knowledge distillation (KD) exchanges soft predictions rather than weights and sidesteps this obstacle, yet convergence theory for fully decentralized, asynchronous peer-to-peer (P2P) KD is lacking. We provide one, relocating consensus from parameter space to function (output) space: a KD event is a geometric contraction operator in logit space on the peers' predictive distributions, which we analyse in the Hilbert space of predictions on a reference measure. Under standard smoothness/variance assumptions and two realizability assumptions, one bridging parameter SGD to the functional step and one controlling restricted task/KD alignment, the time-averaged functional stationarity and function-space disagreement converge at rate $O(1/(ηT))$ to an $O(η)+O(B_f^2)+O(ζ_f^2)$ neighbourhood. Here $B_f$ is the distance from the task optimum to the peers' reachable classes and $ζ_f$ measures persistent local-task heterogeneity. Across homogeneous, width-heterogeneous, and mixed-family networks of the experiments, KD contracts function disagreement by $40-61\times$, while isolated training does not. The sampled stationarity diagnostic has late transient exponents $0.99-1.90$ on the shared-skeleton main runs, and the four-point step-size sweep exhibits the predicted transient: neighbourhood tradeoff.

cs.LG↗

arXiv2026

Sell More, Play Less: Benchmarking LLM Realistic Selling Skill

Sales dialogues require multi-turn, goal-directed persuasion under asymmetric incentives, which makes them a challenging setting for large language models (LLMs). Yet existing dialogue benchmarks rarely measure deal progression and outcomes. We introduce SalesLLM benchmark, a bilingual (ZH/EN) benchmark derived from realistic applications covering Financial Services and Consumer Goods, built from 30,074 scripted configurations and 1,805 curated multi-turn scenarios with controllable difficulty and personas. We propose a fully automatic evaluation pipeline that combines (i) an LLM judge for sales-process progress, and (ii) fine-tuned BERT classifiers for end-of-dialogue buying intent. To improve simulation fidelity, we train a user model, CustomerLM, with SFT and DPO on 8,000+ crowdworker-involved sales conversations, reducing role inversion from 17.44% (GPT-4o) to 8.8%. SalesLLM benchmark scores correlate strongly with human ratings (average Pearson r=0.86; inter-annotator agreement Krippendorff's alpha=0.86, 500 annotations per model). Across 15 mainstream LLMs, in Chinese the strongest models are competitive with typical (junior-to-intermediate) human salespeople -- not with sales experts -- while weaker ones fall below this baseline, and cross-lingual consistency remains poor. SalesLLM benchmark serves as a scalable benchmark for outcome-oriented sales agents. Our code and data are released at https://github.com/Bairong-Xdynamics/Benchmarking-LLM-Realistic-Selling-Skill

cs.CL↗

arXiv2026

MemMark: State-Evolution Attribution Watermarking for Agent Long-Term Memory Systems

Memory-backed agents need provenance that can survive leaked or migrated snapshots, where logs, visible outputs, and trusted metadata may be absent. We propose MemMark, a state-evolution attribution watermark that embeds an owner-controlled signal into latent memory-write decisions. At each internal LLM call, MemMark samples among admissible candidates using keyed, distribution-preserving selection, and records cryptographic commitments with signed session anchors and reveal evidence. This makes attribution depend on reproducible backend behavior rather than mutable provenance fields. Across A-Mem and Graphiti on LoCoMo, with three LLM backbones, MemMark preserves memory utility: Overall F1 retains 99.6% of the unwatermarked baseline, while BLEU-1 changes by +0.2%. It also provides usable carrier capacity, with 1.16, 1.14, and 1.26 bits of mean entropy for update-target, link-target, and semantic-realization decisions. In the snapshot-only R3 setting, MemMark recovers the full 40-bit payload from final snapshots, while wrong-key verification remains near chance. Under nine memory-lifecycle attacks, verification distinguishes tampering, evidence deletion, and partial payload recovery. These results show that robust snapshot-only attribution is feasible for long-term agent memory without surviving traces, trusted metadata, or utility-degrading.

cs.CR↗

arXiv2026

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

Macaron-V1 is an open agent-model family for experiential intelligence: learning from experience in real environments and continuing to learn after deployment. It is organized around two system goals. Adaptation is pursued through recursive improvement of versioned model-harness pairs, where experience from one configuration is evaluated under an external contract and used to construct its successor. Collaboration is pursued via the Mixture-of-LoRA (MoL) architecture that freezes a base model, composes specialist LoRA adapters, and selects one LoRA per user turn. The flagship Macaron-V1-Venti (748B) combines a 744B GLM-5.2 base with four LoRAs for chat, agent, coding, and GenUI; the Qwen3.6-35B-based Macaron-V1-Tall (50B) uses the same design for local deployment. This report presents Macaron-V1 as a co-designed system spanning architecture, algorithms, and infrastructure. The MoL architecture supports continual learning through extensible LoRA specialists. The algorithm combines Model-Harness Co-design and recursive self-improvement loop, including the UI4A component-native GenUI harness, a stateful action substrate, versioned Harness Context Protocol contract, and the agentic RL framework MindForge. The supporting infrastructure includes the post-training platform MinT, the long-context RL method LongStraw, and stability techniques for sparse MoE and DSA base models. We evaluate Macaron-V1 on Personal Intelligence, GenUI, and general capability benchmarks against frontier baselines. Our results validate the current system, while compounding gains from continual learning and collective intelligence remain open questions.

cs.LG↗

arXiv2026

Success Leaves Detours: Learning Executable Walkthroughs for Long-Horizon Agents

Test-time self-evolving agents improve by reusing past experience, yet sparse-reward trajectories contain failures, loops, and detours, while summaries often omit the state conditions and action dependencies needed for execution. We study executable Walkthrough induction from sparse-reward trajectories: extracting compact, state-conditioned, and verifiable procedures. Our key observation is that delayed credit identifies actions associated with progress but cannot determine whether they produce facts required by later actions. We propose Trace, a credit-guided, dependency-grounded framework that compiles noisy trajectories into executable Walkthrough Memory. It detects progress anchors from rewards and persistent state changes, propagates credit to identify valuable transitions, and estimates action prerequisites from cross-episode success and failure evidence. Backward dependency slicing then traces required facts to their producers, extracting dependency-consistent action chains while removing irrelevant loops and detours. The resulting Walkthroughs encode entry conditions, ordered state--action--effect steps, and completion and failure predicates, supporting reuse, intermediate-state resumption, and programmatic verification. Experiments on J-TTL, WebShop, and ScienceWorld with three open-source LLMs show that Trace consistently outperforms eight test-time learning and memory baselines. Compared with the strongest baseline, it improves average AUC and Final-$3$ by $30.0%$ and $40.5%$, respectively, while using fewer inference tokens. These results show that long-horizon interaction benefits more from state-conditioned executable procedures than from complete trajectories or abstract summaries.

cs.LG↗

arXiv2026

LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget

Long-context RL post-training is constrained by the lifetime of state and gradients, not attention cost alone. In GRPO, one multi-million-token prompt must serve old-policy and reference scoring plus multiple policy responses, while conventional autograd keeps the prompt graph and all response graphs live alongside model weights, caches, and distributed communication buffers. We present LongStraw, an objective-aware, architecture-aware system for resident-state virtualization, response replay, and distributed-gradient execution. Its transaction captures the shared prompt without autograd, retains only the architecture-required state on explicitly owned pages, restores that state for each group member, scores old/reference branches without a graph, replays one policy response at a time with autograd, and accumulates the resulting gradients before one distributed finalization and optimizer step. This schedule bounds the live training graph by the response suffix while reusing the expensive prompt computation across the complete GRPO group. We instantiate this design for two incompatible model structures. Qwen3.6-27B combines 48 recurrent GDN layers with 16 full-attention layers; LongStraw keeps the compact recurrent state and physically CP8-sharded KV pages, composes global attention through cross-rank LSE/output merging, and performs blockwise response replay. GLM-5.2 combines a 78-layer MLA/DSA attention stack with a 256-expert, top-8 MoE tail. Its implementation keeps CP-sharded MLA latent pages and DSA indexer-key pages in CPU memory, stages one layer at a time, reconstructs IndexShare-aware global sparse selection over CP32, and dispatches routed response tokens over EP32. The two paths share one transaction contract while specializing the retained state, replay operator, and collective communication to the architecture...

cs.LG↗

arXiv2026

SuperFlow: Training Flow Matching Models with RL on the Fly

Recent progress in flow-based generative models and reinforcement learning (RL) has improved text-image alignment and visual quality. However, current RL training for flow models still has two main problems: (i) GRPO-style fixed per-prompt group sizes ignore variation in sampling importance across prompts, which leads to inefficient sampling and slower training; and (ii) trajectory-level advantages are reused as per-step estimates, which biases credit assignment along the flow. We propose SuperFlow, an RL training framework for flow-based models that adjusts group sizes with variance-aware sampling and computes step-level advantages in a way that is consistent with continuous-time flow dynamics. Empirically, SuperFlow reaches promising performance while using only 5.4% to 56.3% of the original training steps and reduces training time by 5.2% to 16.7% without any architectural changes. On standard text-to-image (T2I) tasks, including text rendering, compositional image generation, and human preference alignment, SuperFlow improves over SD3.5-M by 4.6% to 47.2%, and over Flow-GRPO by 1.7% to 16.0%.

cs.CV↗

arXiv2026

MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs

Evaluating the robustness of Large Vision-Language Models (LVLMs) is essential for their continued development and responsible deployment in real-world applications. However, existing robustness benchmarks typically focus on hallucination or misleading textual inputs, while largely overlooking the equally critical challenge posed by misleading visual inputs in assessing visual understanding. To fill this important gap, we introduce MVI-Bench, the first comprehensive benchmark specially designed for evaluating how Misleading Visual Inputs undermine the robustness of LVLMs. Grounded in fundamental visual primitives, the design of MVI-Bench centers on three hierarchical levels of misleading visual inputs: Visual Concept, Visual Attribute, and Visual Relationship. Using this taxonomy, we curate six representative categories and compile 1,248 expertly annotated VQA instances. To facilitate fine-grained robustness evaluation, we further introduce MVI-Sensitivity, a novel metric that characterizes LVLM robustness at a granular level. Empirical results across 18 state-of-the-art LVLMs uncover pronounced vulnerabilities to misleading visual inputs, and our in-depth analyses on MVI-Bench provide actionable insights that can guide the development of more reliable and robust LVLMs. The benchmark and codebase can be accessed at https://github.com/chenyil6/MVI-Bench.

cs.CV↗

arXiv2026

CoBind: Stage-Aware Compositional Binding for Training-Free Text-to-Image Generation

Diffusion-based text-to-image models often fail on complex prompts involving multiple entities, attributes, and relations, producing object omissions, incorrect attribute assignments, or reversed spatial layouts. Existing training-free methods mainly strengthen token-level attention, but do not explicitly model which attributes belong to which entities or when different constraints should be enforced during denoising. We introduce \textbf{CoBind}, a training-free framework for stage-aware compositional binding. CoBind parses a prompt into a composition graph of entities, attributes, and relations. It first establishes the global layout using entity-completeness and relation constraints, then binds attributes to their target entities through contrastive cross-entity optimization. Structural guidance is gradually relaxed in later denoising steps to preserve textures and visual details. CoBind also adapts the guidance strength according to the current satisfaction of each constraint, reducing unnecessary latent updates. CoBind requires no retraining or additional annotations. Experiments on T2I-CompBench++, GenEval, and multiple diffusion backbones show consistent improvements in attribute binding, spatial relations, and complex compositional generation while maintaining competitive visual quality.

cs.CV↗

arXiv2026

FedCausal-Dyn: A Causal-Dynamic Paradigm for Federated Learning under Dynamic Feature Drift

This paper addresses the challenging problem of dynamic feature drift in federated learning, where data distributions evolve across clients and over time -- a common scenario in real-world applications like financial technology. Existing approaches often assume static drift, limiting their effectiveness in non-stationary environments. To overcome this, we propose \textbf{FedCausal-Dyn}, a novel federated learning framework built on a causal-dynamic paradigm. Its key innovation is \textit{causal-domain feature separation}, which disentangles domain-invariant causal features from spurious, domain-specific variations via specialized projection heads and adversarial training. This enables \textit{reliable and dynamic prototype aggregation}, weighting local class prototypes by estimated reliability before global aggregation. We further introduce \textit{causal-feature guided collaborative regularization}, unifying prototype contrastive alignment and domain invariance into a cohesive objective. Extensive experiments on three federated domain generalization benchmarks demonstrate that FedCausal-Dyn consistently achieves state-of-the-art performance, with the highest average accuracy and the most stable results. Ablation studies confirm each component's critical contribution. Our work provides a robust and principled solution for federated learning under dynamic feature drift.

cs.LG↗

arXiv2026

STRIDE: Strategic Trajectory Reasoning via Discriminative Estimation for Verifiable Reinforcement Learning

Reinforcement Learning with Verifiable Rewards (RLVR) has become an effective post-training paradigm for improving the reasoning abilities of large language models. However, existing RLVR methods typically rely on final-answer correctness to assign trajectory-level rewards, providing sparse supervision and treating all tokens uniformly regardless of their actual contribution to reasoning. Although recent studies introduce intermediate signals such as process rewards, high-entropy tokens, and semantic uncertainty, these signals are often not inherently verifiable and may fail to distinguish beneficial strategic patterns from harmful ones. To address this limitation, we propose STRIDE (Strategic Trajectory Reasoning with Discriminative Estimation), a fine-grained RLVR framework that derives strategic reasoning supervision from verifiable outcomes. STRIDE contrasts successful and failed trajectories within each response group to estimate the outcome-discriminative preference of each $n$-gram strategic pattern, and further combines this signal with reasoning saliency entropy to identify decision-relevant strategic patterns. These patterns are assigned differentiated advantage values during RL optimization, enabling more precise credit assignment while preserving the verifiability of RLVR. Extensive experiments demonstrate that STRIDE consistently improves reasoning performance across diverse models, tasks, and extended settings, including VLMs and agent-based systems.

cs.AI↗

arXiv2026

On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters

Parameter-efficient fine-tuning (PEFT) is usually treated as a cheaper alternative to full fine-tuning. We study a broader role: small trainable adapters as persistent local state on top of strong shared foundation models. In this framing, the base model provides shared competence while adapters carry instance-specific behavior such as preferences, skills, tool habits, and memory-like updates. We organize the problem around three scaling axes: Scale Up, where stronger shared priors make small local updates more useful; Scale Down, where we study how small adapters can be while remaining reliable; and Scale Out, where many persistent adapted instances coexist. MinT provides one infrastructure example for managing adapter identity, revision, provenance, evaluation, and serving residency. Together, the results suggest that PEFT can be a compact substrate for persistent personal models rather than only a budget substitute for full fine-tuning.

cs.LG↗

arXiv2026

GUI Agents for Continual Game Generation

Generating a game is not the same as making one that can be played. Despite advances in code generation, existing approaches treat game generation as one-shot translation from prompt to artifact, leaving interaction-level failures undetected. We argue that evaluating and improving game generation requires a player, and study two roles for graphical user interface (GUI) agents in this process: (1) as an objective evaluator, for which we introduce PlaytestArena, a new evaluation environment that pairs 200 browser-based game generation tasks across eight genres with rubrics of expected in-play behaviors, adjudicated by a GUI agent that loads each build in a browser and plays it; and (2) as a subjective playtester, for which we propose Play2Code, where a game agent and a GUI agent operate in a sustained loop with shared memory, turning game generation into a dialogue between coding and playing. Our experiments show that even frontier models struggle to generate playable games directly, while Play2Code achieves a 66.8\% rubric pass-rate, improving over single-pass and agentic-coding baselines by 37.1 and 14.6 points respectively. Further analysis shows that GUI playtester feedback is more traceable than a human report, yet idiosyncratic in ways reminiscent of human testers, establishing game playtesting as a critical testbed for interactive code generation. Our project website is available at https://continual-game-generation.vercel.app/.

cs.SE↗

arXiv2026

MinT: Managed Infrastructure for Training and Serving Millions of LLMs

We present MindLab Toolkit (MinT), a managed infrastructure system for Low-Rank Adaptation (LoRA) post-training and online serving. MinT targets a setting where many trained policies are produced over a small number of expensive base-model deployments. Instead of materializing each policy as a merged full checkpoint, MinT keeps the base model resident and moves exported LoRA adapter revisions through rollout, update, export, evaluation, serving, and rollback, hiding distributed training, serving, scheduling, and data movement behind a service interface. MinT scales this path along three axes. Scale Up extends LoRA RL to frontier-scale dense and MoE architectures, including MLA and DSA attention paths, with training and serving validated beyond 1T total parameters. Scale Down moves only the exported LoRA adapter, which can be under 1% of base-model size in rank-1 settings; adapter-only handoff reduces the measured step by 18.3x on a 4B dense model and 2.85x on a 30B MoE, while concurrent multi-policy GRPO shortens wall time by 1.77x and 1.45x without raising peak memory. Scale Out separates durable policy addressability from CPU/GPU working sets: a tensor-parallel deployment supports 10^6-scale addressable catalogs (measured single-engine sweeps through 100K) and thousand-adapter active waves at cluster scale, with cold loading treated as scheduled service work and packed MoE LoRA tensors improving live engine loading by 8.5-8.7x. MinT thus manages million-scale LoRA policy catalogs while training and serving selected adapter revisions over shared 1T-class base models.

cs.LG↗

arXiv2026

Macaron-A2UI: A Model for Generative UI in Personal Agents

As personal agents evolve to handle complex, user-centric tasks, static plain-text chat is rapidly becoming a bottleneck. Generative UI emerges as the necessary new interface layer, dynamically synthesizing the right controls, options, and state from the interaction context in real time. We present Macaron-A2UI, a model for Generative UI in personal agents. Our goal is to move beyond text-only interaction by enabling agents to generate natural language together with lightweight, executable UI actions for information collection, preference refinement, confirmation, and multi-goal organization. We build a large-scale Generative UI corpus from heterogeneous dialogue sources, introduce A2UI-Bench for controlled evaluation, and train 30B, 235B and 754B models with parameter-efficient LoRA-based supervised fine-tuning followed by reward-driven reinforcement learning. The best Macaron-A2UI model reaches 75.6 overall on A2UI-Bench without explicit schema hints, surpassing the strongest full-schema frontier baseline. We release the models, benchmark, and evaluation protocol to support future work on Generative UI for personal agents.

cs.HC↗

arXiv2026

Experimental Study of Bremsstrahlung Gamma Ray Emission and Short-Range Correlations in $^{124}$Sn+$^{124}$Sn Collisions at 25 MeV/u

Short-range correlation (SRC) in nuclei refers to nucleons forming temporally correlated pairs in close proximity, giving rise to the high momentum of the nucleons beyond the Fermi surface. It has been reported that bremsstrahlung $γ$ production from neutron-proton process in heavy-ion reactions provides a potential probe to the SRC abundance in nuclei. In this paper, we present in detail the precision measurement of bremsstrahlung $γ$-rays in $\rm ^{124}Sn$+$\rm ^{124}Sn$ reactions at 25 MeV/u using the Compact Spectrometer for Heavy IoN Experiment (CSHINE). A comprehensive experimental and analysis framework is established to ensure the reliability and robustness of the extracted results. Background contributions are evaluated and subtracted using independent methods, and the consistency of the analysis is systematically validated. By comparing the experimental $γ$ spectrum with the Isospin-dependent Boltzmann-Uehling-Uhlenbeck simulations, the high momentum tail (HMT) fraction of $R_{\rm HMT}=(20 \pm 3)\%$ is derived in $^{124}$Sn nuclei. This work provides a detailed and validated experimental framework for extracting SRC information from bremsstrahlung $γ$-ray emission and demonstrates the feasibility of studying nucleon SRCs with high precision in low-energy heavy-ion collisions.

nucl-ex↗

arXiv2026

High-Precision Measurement of D($γ$, $n$)$p$ Photodisintegration Reaction and Implications for Big-Bang Nucleosynthesis

We report on a high-precision measurement of the D($γ$,\,$n$)$p$ photodisintegration reaction at the newly commissioned Shanghai Laser Electron Gamma Source (SLEGS), employing a quasi-monochromatic $γ$-ray beam from Laser Compton Scattering. The cross sections were determined over $E_γ$=2.327--7.089 MeV, achieving up to a factor of 2.2 improvement in precision near the neutron separation threshold. Combined with previous data in a global Markov chain Monte Carlo (MCMC) analysis using dibaryon effective field theory, we obtained the unprecedentedly precise $p$($n$,\,$γ$)D cross sections and thermonuclear rate, with a precision up to $\approx$4 times higher than previous evaluations. Implemented in a standard Big-Bang Nucleosynthesis (BBN) framework, this new rate decreases uncertainty of the key cosmological parameter of baryon density $Ω_b h^2$ by up to $\approx$16\% relative to the LUNA result. A residual $\approx$1.2$σ$ tension between $Ω_b h^2$ constrained from primordial D/H observations and CMB measurements persists, highlighting the need for improved $dd$ reaction rates and offering potential hints of new physics beyond the standard model of cosmology.

nucl-ex↗

arXiv2025

Precise Measurement of Short-Range Correlations in Nuclei from Bremsstrahlung Gamma Ray Emission in Low-Energy Heavy-Ion Collisions

Atomic nuclei and dense nucleonic matter in neutron stars exhibit short-range correlations (SRCs), where nucleons form temporally correlated pairs in proximity beyond mean-field approximation. It is essential to make precision measurement of the fraction of SRC since it carries the signature of underlying quark dynamics in nuclear medium. In this letter, we present the first high-precision measurement of neutron-proton bremsstrahlung $γ$-ray emission from the symmetric $\rm ^{124}Sn$+$\rm ^{124}Sn$ reactions at 25 MeV/u. From the observed spectral hardening, the precise SRC fraction in the $\rm ^{124}Sn$ nucleus is extracted to be $(20 \pm 3)\%$. This result provides a novel, direct and unambiguous evidence of SRCs, and demonstrates that low-energy heavy-ion collisions offers a new approach to studying nuclear structure in connection with quark-level dynamics.

nucl-ex↗