Searcharxiv⌕ Search

arXiv subjects

Jiajun Shen

Publications and source records attributed to Jiajun Shen.

At least 19 recordsLinked to original sources

MemT-1.0: Towards Memory-Native Evolution

Memory-native evolution requires making memory an intrinsic capability of foundation models, enabling them to continuously accumulate and utilize information. To this end, we introduce MemT-1.0, a new architecture of memory foundation models that internalize memory as part of model computation. Building on Metis, MemT-1.0 employs Gated DeltaNet-2 with a stable token-level memory update mechanism within each turn. We further introduce Mixture-of-Storage to scale native memory capacity. Following a deliberate backbone preparation, we develop a two-stage training pipeline consisting of continued pre-training and post-training, supported by 34B tokens of training data and efficient training and inference infrastructure. We release a model family comprising MemT-1.0-Nano, MemT-1.0-Flash, and MemT-1.0-Pro, and conduct a comprehensive evaluation of memory and general capabilities. The results show that they outperform the parametric memory baselines on average and remain competitive with retrieval-augmented methods, while largely preserving general capabilities. At a history length of 128K tokens, MemT-1.0-Nano achieves approximately $1.86\times$ end-to-end inference speedup and a $66\times$ reduction in persistent storage relative to full context KV cache. By internalizing retrieved procedural experience, MemT-1.0 improves task success on agentic tasks, providing a practical step toward memory-native evolution.

cs.CL↗

Assessing the CAD Capabilities of General-Purpose AI Models: A Case Study of GPT-6 Astra

Recent advances in general-purpose AI models and agentic systems have enabled them to perform some CAD tasks. A systematic assessment is now needed to establish what these systems can accomplish and how well their outputs meet engineering requirements. This paper presents an empirical study of GPT-6 Astra, a SOTA general-purpose AI model, across four basic CAD tasks: model understanding, generation, editing, and evaluation. We design test cases spanning different levels of complexity and engineering requirements. The results show that GPT-6 Astra can support multiple stages of CAD workflows, with successful outcomes in cases involving nontrivial geometric complexity. This suggests that general-purpose AI is beginning to offer practical value in almost all steps in CAD workflow. However, gaps remain between generating visually plausible geometry and producing valid, functional, and editable CAD models. Recurring errors in freeform geometry, parameter dependencies, and mating relationships between parts limit engineering usefulness. Agentic iterative refinement can improve results but does not consistently resolve critical errors; long completion times further constrain practical use.

cs.GR↗

Model Distribution-Aware Multimodal Dataset Distillation

Multimodal dataset distillation (MDD) aims to reduce the cost of vision-language pre-training by synthesizing compact image-text datasets. Although recent feature-matching methods have improved distillation efficiency, synthetic data may become specialized to the model representations encountered during optimization, limiting generalization to unseen model states. We propose Model Distribution-Aware Multimodal Dataset Distillation (MDA-DD), which leverages supervision from a dynamic staggered queue of models at different training stages. We further introduce asymmetric dual projection for complementary alignment in visual and linguistic spaces and an uncentered second-order moment objective that retains mean-related cross-modal information. Experiments on Flickr30K and MS-COCO demonstrate consistent retrieval improvements over existing methods across synthetic-data budgets. With only 100 synthetic pairs, MDA-DD improves IR@1 and TR@1 on Flickr30K by 12.9 and 18.2 percentage points, respectively, over the strongest baseline at the same budget, outperforming all compared methods using 500 pairs on both datasets across all six retrieval metrics.

cs.CV↗

ABot-ER: Unified Embodied Reasoning through Agentic Data-Model Co-Evolution

We introduce ABot Embodied Reasoning (ABot-ER), a unified embodied reasoning foundation model that breaks the tension between breadth and depth in embodied reasoning. Embodied intelligence demands the joint mastery of perception, planning, and execution, yet reconciling generality with specialization across this spectrum remains an open challenge. Within a single unified model, ABot-ER commands specialist-level depth across all three competencies, spanning spatial, scene, and temporal reasoning in perception; long-horizon action organization in planning; and point grounding and affordance in execution. The Agentic Data-Model Self-Evolution Engine continuously adapts data strategies to the model's evolving capability state. The Task-Conditioned Adaptive Training Strategy uses task tokens to activate lightweight low-rank residuals over a shared backbone, separating updates from heterogeneous output formats while retaining shared embodied knowledge. Across a systematic evaluation on twelve benchmarks covering three core competencies and five capability dimensions, ABot-ER achieves state-of-the-art overall performance across perception, planning, and execution within a unified parametric framework, demonstrating that breadth and depth can coexist in unified embodied reasoning.

cs.AI↗

PHASE-Tree: Modeling Character-State Evolution in Long-Horizon Role-Playing Dialogue

Long-horizon role-playing demands that characters remain recognizable as they evolve with the narrative. Yet existing work falls short on two fronts: representations are typically static profiles that cannot be updated locally without destabilizing unchanged traits, and benchmarks mainly test persona preservation and memory recall rather than whether a model speaks from a character's currently evolved state. We address both. PHASE-Tree is a multi-timescale character-state tree with an immutable identity root and mutable persona, session, and moment layers, making each mutable field an addressable target for localized within- and cross-episode updates. It conditions generation through explicit textual provision or implicit parametric adaptation. To measure evolved-state generation, we introduce LongEvoRoleBench, which pairs four long-dialogue corpora for cross-episode evolution with four short-dialogue corpora as within-scene state-tracking checks, under a unified next-utterance protocol. On the long-dialogue core, textual PHASE-Tree ranks first in 11 of 12 dataset-metric cells against internal variants and all 12 cells against external textual baselines, improving character-level, semantic, and embedding scores by 19.7%, 12.4%, and 15.1% respectively. In a blinded 200-response study, human ratings correlate with the GPT-4.1 judge (Pearson r= 0.65); on descriptive n= 10 PT and NR prompt subsets, the Overall difference is +0.20. The long-dialogue Sem advantage persists across LLM judges and generation backbones.

cs.CL↗

SkillSmith: Learning to Compose Parametric Skills and Textual Knowledge

Agentic systems driven by large language models (LLMs) regularly feature two key mechanisms to autonomously solve complex problems: synthesizing text-based knowledge and procedures from past experiences and building parametric (weight-space) skill libraries for recurring sub-goals. To date, research has largely treated these as orthogonal pursuits: either organizing textual knowledge through composition and reflection, or consolidating parametric skills via weight-space merging. Consequently, the seamless integration of text and model weights for targeted performance improvements remains largely unexplored. This work bridges this modality gap by treating model weights as an additional modality that an LLM can natively reason over. We instantiate parametric learning via prefix-tuning and augment an LLM to ingest both prefix weights and rich textual data which capture relationships to a target capability. Our augmented LLM, which we call SkillSmith, synthesizes these inputs to perform instruction-steered parametric synthesis, directly outputting new prefix weights that manifest the target skill. We demonstrate that our approach significantly outperforms both text-only and weight-space-only baselines, unlocking performance gains that are out of reach for uni-modal (text-only or weight-only) adaptations.

cs.CL↗

MemReranker: Reasoning-Aware Reranking for Agent Memory Retrieval

In agent memory systems, the reranking model serves as the critical bridge connecting user queries with long-term memory. Most systems adopt the "retrieve-then-rerank" two-stage paradigm, but generic reranking models rely on semantic similarity matching and lack genuine reasoning capabilities, leading to a problem where recalled results are semantically highly relevant yet do not contain the key information needed to answer the question. This deficiency manifests in memory scenarios as three specific problems. First, relevance scores are miscalibrated, making threshold-based filtering difficult. Second, ranking degrades when facing temporal constraints, causal reasoning, and other complex queries. Third, the model cannot leverage dialogue context for semantic disambiguation. This report introduces MemReranker, a reranking model family (0.6B/4B) built on Qwen3-Reranker through multi-stage LLM knowledge distillation. Multi-teacher pairwise comparisons generate calibrated soft labels, BCE pointwise distillation establishes well-distributed scores, and InfoNCE contrastive learning enhances hard-sample discrimination. Training data combines general corpora with memory-specific multi-turn dialogue data covering temporal constraints, causal reasoning, and coreference resolution. On the memory retrieval benchmark, MemReranker-0.6B substantially outperforms BGE-Reranker and matches open-source 4B/8B models as well as GPT-4o-mini on key metrics. MemReranker-4B further achieves 0.737 MAP, with several metrics on par with Gemini-3-Flash, while maintaining inference latency at only 10--20% of large models. On finance and healthcare vertical-domain benchmarks, the models preserve generalization capabilities on par with mainstream large-parameter rerankers.

cs.CL↗

Context Training with Active Information Seeking

Most existing large language models (LLMs) are expensive to adapt after deployment, especially when a task requires newly produced information or niche domain knowledge. Recent work has shown that, by manipulating and optimizing their context, LLMs can be tailored to downstream tasks without updating their weights. However, most existing methods remain closed-loop, relying solely on the model's intrinsic knowledge. In this paper, we equip these context optimizers with Wikipedia search and browser tools for active information seeking. We show that naively adding these tools to a standard sequential context optimization pipeline can actually degrade performance compared to baselines. However, when paired with a search-based training procedure that maintains and prunes multiple candidate contexts, active information seeking delivers consistent and substantial gains. We demonstrate these improvements across diverse domains, including low-resource translation (Flores+), health scenarios (HealthBench), and reasoning-heavy tasks (LiveCodeBench and Humanity's Last Exam). Furthermore, our method proves to be data-efficient, robust across different hyperparameters, and capable of generating effective textual contexts that generalize well across different models.

cs.CL↗

COMBA: Cross Batch Aggregation for Learning Large Graphs with Context Gating State Space Models

State space models (SSMs) have recently emerged for modeling long-range dependency in sequence data, with much simplified computational costs than modern alternatives, such as transformers. Advancing SMMs to graph structured data, especially for large graphs, is a significant challenge because SSMs are sequence models and the shear graph volumes make it very expensive to convert graphs as sequences for effective learning. In this paper, we propose COMBA to tackle large graph learning using state space models, with two key innovations: graph context gating and cross batch aggregation. Graph context refers to different hops of neighborhood for each node, and graph context gating allows COMBA to use such context to learn best control of neighbor aggregation. For each graph context, COMBA samples nodes as batches, and train a graph neural network (GNN), with information being aggregated cross batches, allowing COMBA to scale to large graphs. Our theoretical study asserts that cross-batch aggregation guarantees lower error than training GNN without aggregation. Experiments on benchmark networks demonstrate significant performance gains compared to baseline approaches. Code and benchmark datasets will be released for public access.

cs.LG↗

Latent Space Communication via K-V Cache Alignment

Solving increasingly complex problems with large language models (LLMs) necessitates a move beyond individual models and towards multi-model systems that can effectively collaborate. While text has traditionally served as the medium for inter-model communication, a richer and more efficient exchange is possible if models can access each other's internal states directly. In this paper, we propose learning a shared representation space that aligns the k-v caches of multiple models, creating a high-bandwidth channel for collaboration without altering the underlying pre-trained parameters. We do so by augmenting each model with adapters to translate its state into and out of this shared space. Via a suite of experiments with Gemma-2 models, we demonstrate that this approach not only enables seamless inter-model communication but also improves individual model performance. We also show that the shared space allows for the direct transfer of learned skills, such as soft prompts, between different models. Our work represents a significant step towards a future where models can fluidly share knowledge and capabilities.

cs.LG↗

Composite Learning Adaptive Control under Non-Persistent Partial Excitation

This paper focuses on relaxing the excitation conditions for the adaptive control of uncertain nonlinear systems. By adopting the spectral decomposition technique, a linear regression equation (LRE) is constructed to quantitatively collect historical excitation information, based on which the parameter estimation error is decomposed into the excited component and the unexcited component. By sufficiently utilizing the collected excitation information, the composite learning and μ-modification terms are designed and incorporated into the "Lyapunov-based" parameter update law. By developing a novel Lyapunov function, it is demonstrated that under non-persistent partial excitation, the control error and the excited parameter estimation error component converge to zero, while the unexcited component remains bounded. Furthermore, the proposed adaptive control scheme can effectively eliminate the effects of parametric uncertainties and enhance the robustness of the closed-loop systems. Simulation results are provided to verify the theoretical findings.

eess.SY↗

LHGEL: Large Heterogeneous Graph Ensemble Learning using Batch View Aggregation

Learning from large heterogeneous graphs presents significant challenges due to the scale of networks, heterogeneity in node and edge types, variations in nodal features, and complex local neighborhood structures. This paper advocates for ensemble learning as a natural solution to this problem, whereby training multiple graph learners under distinct sampling conditions, the ensemble inherently captures different aspects of graph heterogeneity. Yet, the crux lies in combining these learners to meet global optimization objective while maintaining computational efficiency on large-scale graphs. In response, we propose LHGEL, an ensemble framework that addresses these challenges through batch sampling with three key components, namely batch view aggregation, residual attention, and diversity regularization. Specifically, batch view aggregation samples subgraphs and forms multiple graph views, while residual attention adaptively weights the contributions of these views to guide node embeddings toward informative subgraphs, thereby improving the accuracy of base learners. Diversity regularization encourages representational disparity across embedding matrices derived from different views, promoting model diversity and ensemble robustness. Our theoretical study demonstrates that residual attention mitigates gradient vanishing issues commonly faced in ensemble learning. Empirical results on five real heterogeneous networks validate that our LHGEL approach consistently outperforms its state-of-the-art competitors by substantial margin. Codes and datasets are available at https://github.com/Chrisshen12/LHGEL.

cs.LG↗

StrucADT: Generating Structure-controlled 3D Point Clouds with Adjacency Diffusion Transformer

In the field of 3D point cloud generation, numerous 3D generative models have demonstrated the ability to generate diverse and realistic 3D shapes. However, the majority of these approaches struggle to generate controllable 3D point cloud shapes that meet user-specific requirements, hindering the large-scale application of 3D point cloud generation. To address the challenge of lacking control in 3D point cloud generation, we are the first to propose controlling the generation of point clouds by shape structures that comprise part existences and part adjacency relationships. We manually annotate the adjacency relationships between the segmented parts of point cloud shapes, thereby constructing a StructureGraph representation. Based on this StructureGraph representation, we introduce StrucADT, a novel structure-controllable point cloud generation model, which consists of StructureGraphNet module to extract structure-aware latent features, cCNF Prior module to learn the distribution of the latent features controlled by the part adjacency, and Diffusion Transformer module conditioned on the latent features and part adjacency to generate structure-consistent point cloud shapes. Experimental results demonstrate that our structure-controllable 3D point cloud generation method produces high-quality and diverse point cloud shapes, enabling the generation of controllable point clouds based on user-specified shape structures and achieving state-of-the-art performance in controllable point cloud generation on the ShapeNet dataset.

cs.GR↗

HGEN: Heterogeneous Graph Ensemble Networks

This paper presents HGEN that pioneers ensemble learning for heterogeneous graphs. We argue that the heterogeneity in node types, nodal features, and local neighborhood topology poses significant challenges for ensemble learning, particularly in accommodating diverse graph learners. Our HGEN framework ensembles multiple learners through a meta-path and transformation-based optimization pipeline to uplift classification accuracy. Specifically, HGEN uses meta-path combined with random dropping to create Allele Graph Neural Networks (GNNs), whereby the base graph learners are trained and aligned for later ensembling. To ensure effective ensemble learning, HGEN presents two key components: 1) a residual-attention mechanism to calibrate allele GNNs of different meta-paths, thereby enforcing node embeddings to focus on more informative graphs to improve base learner accuracy, and 2) a correlation-regularization term to enlarge the disparity among embedding matrices generated from different meta-paths, thereby enriching base learner diversity. We analyze the convergence of HGEN and attest its higher regularization magnitude over simple voting. Experiments on five heterogeneous networks validate that HGEN consistently outperforms its state-of-the-art competitors by substantial margin.

cs.LG↗

Dual State-space Fidelity Blade (D-STAB): A Novel Stealthy Cyber-physical Attack Paradigm

This paper presents a novel cyber-physical attack paradigm, termed the Dual State-Space Fidelity Blade (D-STAB), which targets the firmware of core cyber-physical components as a new class of attack surfaces. The D-STAB attack exploits the information asymmetry caused by the fidelity gap between high-fidelity and low-fidelity physical models in cyber-physical systems. By designing precise adversarial constraints based on high-fidelity state-space information, the attack induces deviations in high-fidelity states that remain undetected by defenders relying on low-fidelity observations. The effectiveness of D-STAB is demonstrated through a case study in cyber-physical battery systems, specifically in an optimal charging task governed by a Battery Management System (BMS).

eess.SY↗

Transparentize the Internal and External Knowledge Utilization in LLMs with Trustworthy Citation

While hallucinations of large language models could been alleviated through retrieval-augmented generation and citation generation, how the model utilizes internal knowledge is still opaque, and the trustworthiness of its generated answers remains questionable. In this work, we introduce Context-Prior Augmented Citation Generation task, requiring models to generate citations considering both external and internal knowledge while providing trustworthy references, with 5 evaluation metrics focusing on 3 aspects: answer helpfulness, citation faithfulness, and trustworthiness. We introduce RAEL, the paradigm for our task, and also design INTRALIGN, an integrated method containing customary data generation and an alignment algorithm. Our experimental results show that our method achieves a better cross-scenario performance with regard to other baselines. Our extended experiments further reveal that retrieval quality, question types, and model knowledge have considerable influence on the trustworthiness in citation generation.

cs.CL↗

Streaming DiLoCo with overlapping communication: Towards a Distributed Free Lunch

Training of large language models (LLMs) is typically distributed across a large number of accelerators to reduce training time. Since internal states and parameter gradients need to be exchanged at each and every single gradient step, all devices need to be co-located using low-latency high-bandwidth communication links to support the required high volume of exchanged bits. Recently, distributed algorithms like DiLoCo have relaxed such co-location constraint: accelerators can be grouped into ``workers'', where synchronizations between workers only occur infrequently. This in turn means that workers can afford being connected by lower bandwidth communication links without affecting learning quality. However, in these methods, communication across workers still requires the same peak bandwidth as before, as the synchronizations require all parameters to be exchanged across all workers. In this paper, we improve DiLoCo in three ways. First, we synchronize only subsets of parameters in sequence, rather than all at once, which greatly reduces peak bandwidth. Second, we allow workers to continue training while synchronizing, which decreases wall clock time. Third, we quantize the data exchanged by workers, which further reduces bandwidth across workers. By properly combining these modifications, we show experimentally that we can distribute training of billion-scale parameters and reach similar quality as before, but reducing required bandwidth by two orders of magnitude.

cs.CL↗

A Simple Recipe for Contrastively Pre-training Video-First Encoders Beyond 16 Frames

Understanding long, real-world videos requires modeling of long-range visual dependencies. To this end, we explore video-first architectures, building on the common paradigm of transferring large-scale, image--text models to video via shallow temporal fusion. However, we expose two limitations to the approach: (1) decreased spatial capabilities, likely due to poor video--language alignment in standard video datasets, and (2) higher memory consumption, bottlenecking the number of frames that can be processed. To mitigate the memory bottleneck, we systematically analyze the memory/accuracy trade-off of various efficient methods: factorized attention, parameter-efficient image-to-video adaptation, input masking, and multi-resolution patchification. Surprisingly, simply masking large portions of the video (up to 75%) during contrastive pre-training proves to be one of the most robust ways to scale encoders to videos up to 4.3 minutes at 1 FPS. Our simple approach for training long video-to-text models, which scales to 1B parameters, does not add new architectural complexity and is able to outperform the popular paradigm of using much larger LLMs as an information aggregator over segment-based information on benchmarks with long-range temporal dependencies (YouCook2, EgoSchema).

cs.CV↗