Searcharxiv⌕ Search

arXiv subjects

Tong Zhao

Publications and source records attributed to Tong Zhao.

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

EGR: Embedding-Native Generative Retrieval with a Shared LLM

Generative retrieval is increasingly popular in large-scale recommendation and advertising systems, yet current methods introduce practical complications. Semantic-ID methods rely on quantization, mutable identifier vocabularies, and token-to-item grounding; embedding-based pipelines train the item encoder separately from the query generator, which limits user-item alignment. We propose EGR, an Embedding-Native Generative Retrieval framework for recommendation and advertising. EGR uses a single shared LLM to learn item representations from item metadata and user representations from interaction histories in one embedding space. Items are indexed directly as dense vectors, and user histories are encoded as dense retrieval queries. Joint contrastive training groups related items and aligns queries with their target items. We evaluate EGR on public benchmarks, industrial data, and live deployment. EGR outperforms published baselines on Amazon Reviews; on Snap DPA, it scales with data, handles cold-start items, and benefits from multimodal input. In production, EGR delivers a +2.91% conversion-rate lift, simplifying system design while improving retrieval quality and ad performance.

cs.IR↗
arXiv2026

Beyond Fixed Depths and Widths: Optimizing Textual Decoding Tries in LLM-based Generative Recommendation

Generative recommendation (GR) is an increasingly popular paradigm in recommender systems, with a prominent line of work using LLMs as autoregressive backbones to predict the next item's term IDs (e.g., titles or keywords). The success of autoregressive generation hinges on constrained beam search over a decoding trie to ensure that generated outputs correspond to valid items. However, current research predominantly focuses on generating more comprehensive term IDs to describe items, while largely neglecting the structural design of the decoding trie formed by these terms. This can lead to a trie that is poorly suited to beam search, which degrades performance. To address this, we examine the effectiveness of term IDs from the perspective of decoding trie optimization. Through empirical and theoretical analyses, we identify two desirable properties for a highly performant trie: (1) adaptive and variable ID length, enabling items with varying semantic richness to be represented by IDs of appropriate lengths, and (2) constrained branching factors, especially at shallow levels, which drastically improves the success rate of constrained beam search. Motivated by these properties, we introduce BONSAI: Branching-Optimized Node Structure for Adaptive Identifiers, a novel framework that co-designs textual term IDs and their underlying decoding trie. BONSAI extracts recommendation-informative words from item metadata and employs a minimum set cover formulation to recursively build a trie that satisfies the above properties. Experiments reveal that BONSAI achieves up to a 21.6% relative improvement over state-of-the-art baselines. Further analyses confirm the crucial role of our proposed properties, and demonstrate their generalizability to be applied to enhance the performance of other term ID methods.

cs.IR↗
arXiv2026

Threshold Differential Attention for Sink-Free, Ultra-Sparse, and Non-Dispersive Language Modeling

Softmax attention struggles with long contexts due to structural limitations: the strict sum-to-one constraint forces attention sinks on irrelevant tokens, and probability mass disperses as sequence lengths increase. We tackle these problems with Threshold Differential Attention (TDA), a sink-free attention mechanism that achieves ultra-sparsity and improved robustness at longer sequence lengths without the computational overhead of projection methods or the performance degradation caused by noise accumulation of standard rectified attention. TDA applies row-wise extreme-value thresholding with a length-dependent gate, retaining only exceedances. Inspired by the differential transformer, TDA also subtracts an inhibitory view to enhance expressivity. Theoretically, we prove that TDA controls the expected number of spurious survivors per row to $O(1)$ and that consensus spurious matches across independent views vanish as context grows. Empirically, TDA produces $>99\%$ exact zeros and eliminates attention sinks while maintaining competitive performance on standard and long-context benchmarks.

cs.LG↗
arXiv2026

Constraining Pulsar Radiative Geometry via Multi-wavelength Modeling

We propose that jointly modeling the thermal X-ray pulse profiles and the polarization position angles offers an effective means of locating the polarization orientation focus of the pulsar's coherent radiation. From the X-ray pulse-profile measurement we constrain the colatitude of the center of the thermal X-ray emission, which corresponds to the center of in-falling particles within the polar cap, while the RVM fitting yields the inclination angle of the focus point of polarization orientations. Thus, consistency between these two independent angle measurements would imply that the RVM fit faithfully recovers the inclination of the plasma flow center, and this center coincides with the polarization orientation focus. Conversely, the discrepancy would suggest that the polarization state of the radio emission changes as it propagates because the evolution of wave modes during wave propagation in the magnetosphere strongly depends on magnetic field orientations.

astro-ph.HE↗
arXiv2026

EP251023a: A fast X-ray transient featuring a magnetar-powered optical internal plateau followed by a steep decay

EP251023a is an extragalactic fast X-ray transient (eFXT) detected solely by EP without a gamma-ray counterpart. The prompt emission consists of a main emission with a duration $T_{90}=292\pm19$ s, followed by a long-lasting tail emission that persists until the observation ends at $T_0+1571$ s. With the upper limit of Konus--Wind, we derived a conservative upper limit on the isotropic gamma-ray energy $E_{γ,\rm{iso}}$ of $5.7 \times 10^{52}$ erg for the main emission phase. A redshift of $z = 2.232\pm0.001$ is identified from strong absorption features in the Keck spectrum, which also indicate a relatively low host-galaxy HI column density. Based on the broadband spectral energy distribution, the late-time light curves show an achromatic plateau, followed by an extremely steep decay with a slope of 3.99 after a break at about 49 ks, which is consistent with a rapidly spinning millisecond magnetar engine. Under the isotropic wind scenario, we obtain the initial period $P_0<2.27$~ms and the magnetic field strength $B_p<8.33\times10^{14}$~G for the magnetar; whereas considering a jet collimation with a typical opening angle of 0.1 rad relaxes these constraints to $P_0<32.15$~ms and $B_p<1.18\times10^{16}$~G. Together with GRB\,070707, EP251023a may represent a rare class of optical magnetar-powered internal plateaus with little external-shock contamination, unlike previous examples detected primarily in X-rays. Future discoveries of similar events will help clarify the relationship between magnetar-powered internal emission observed in the optical band and that detected only in X-rays.

astro-ph.HE↗
arXiv2026

OneRetrieval: Unifying Multi-Branch E-commerce Retrieval with an Editable Generative Model

Industrial e-commerce search serves hundreds of millions of items through a multi-branch retrieval stage fused by hand-tuned merging without joint optimization. Generative retrieval (GR) raises the prospect of collapsing this stage into a single model, yet unification is gated by more than retrieval quality: the inverted-index branch converts below the platform average yet persists because it is almost the only branch where operations can inject a new term within hours without any model update; a one-model substitute must preserve this real-time editability. Existing GR methods structurally lack it: closed-codebook methods fix each slot to a quantized embedding at training, while open-vocabulary methods leave new-term routing to model generalization. We present OneRetrieval, a one-model GR framework built on Keyword-Aligned Encoding (KAE), which ties each identifier position to an interpretable attribute word, pairing competitive recall quality with the editability of the inverted index -- to our knowledge the first editable generative retrieval method. An information-theoretic merging organizes 18 attribute categories into six codebook groups with non-uniform capacity; reserved slots in each codebook can be bound to new words after deployment without retraining; and a four-stage fine-tuning pipeline secures quality and editability jointly. On five million real-traffic requests, OneRetrieval matches the deep recall of the strongest generative baseline, with an intervention hit rate over an order of magnitude above closed-codebook encodings. Online, replacing the inverted-index branch significantly lifts order volume; extending to nearly the entire stage holds conversion while improving CTR. The system is deployed at Kuaishou, serving hundreds of millions of PVs daily.

cs.IR↗
arXiv2026

Can We Hear from Events? Generating Speech from Event Camera

Traditional RGB-based speech generation faces Temporal Granularity Mismatch since fixed camera exposure times inevitably blur the high-frequency articulatory transients essential for rendering emotional speech. To break this ceiling, we propose EventSpeech as a novel text-conditioned framework pioneering the use of neuromorphic events for expressive speech generation, since these microsecond-precise events naturally align with acoustic waveform dynamics. Our architecture integrates a dedicated Event Encoder to model sparse neuromorphic events alongside a multi-scale Audio Encoder featuring a Hierarchical Wavelet Contextualizer (HWC). A bidirectional alignment mechanism seamlessly synchronizes linguistic content and visual dynamics with dense acoustic features. Furthermore, we construct EVT-SPK as the first benchmark comprising large-scale synthetic data and real-world recordings from specialized neuromorphic hardware. Extensive evaluations demonstrate that EventSpeech significantly outperforms current baselines by preserving fine-grained emotions and resisting motion blur to establish a new paradigm for multimodal speech generation. Code and demo are available at https://xrfang-0102.github.io/EventSpeechWeb/.

cs.MM↗
arXiv2026

Budget-Constrained Step-Level Diffusion Caching

Step-level caching accelerates diffusion models by exploiting temporal redundancy across denoising steps. Existing methods make per-step cache decisions using threshold-based heuristics, without directly optimizing for final output quality. As a result, their inference latency varies across inputs and is difficult to control at deployment. In this work, we propose BudCache, which inverts this formulation: rather than letting per-step error thresholds dictate the runtime cost, we fix the compute budget in advance and search for the cache policy that best preserves the final output. To tackle the combinatorial complexity of step selection, we combine Simulated Annealing with deterministic Hill Climbing. This offline search identifies high-quality cache policies within minutes and introduces no online search or thresholding overhead during inference. When the compute budget is very tight, we further introduce cache-aware schedule alignment, which adapts the time discretization to the selected cache policy to reduce cache-induced trajectory mismatch. Experiments on FLUX.1-dev and Wan2.1 show that BudCache achieves better generation quality than heuristic caching baselines under the same inference budgets. Code is available at https://github.com/Westlake-AGI-Lab/BudCache

cs.CV↗
arXiv2026

Harnessing Routing Foresight for Micro-step-level MoE load balancing in RL Post-training

Mixture-of-Experts (MoE) and reinforcement learning (RL) post-training now dominate large language model (LLM) development, yet expert load imbalance remains a critical challenge. Existing load-balancing systems target pre-training by relying on historical step-level statistics. However, these methods fail under the unique workload dynamics of RL post-training: the step-level load is stable, but the tiny batch sizes processed during micro-steps cause severe, high-frequency load fluctuations. We introduce ForeMoE, a micro-step-level load balancing system for MoE RL post-training. Instead of relying on historical statistics, ForeMoE exploits the multi-stage RL pipeline (rollout, recompute, policy update) by using foreseeable routing information from the rollout stage to proactively guide load balancing in the remaining stages. To support frequent per-micro-step reconfiguration, ForeMoE employs a hierarchical planner that decomposes the NP-hard load balancing problem into tractable sub-components, alongside a transfer engine that leverages complementary hardware paths (CPU-assisted and GPU-direct) for overlapped expert transfer. Evaluations on 64 GPUs demonstrate that ForeMoE achieves up to a 1.45$\times$ speedup over state-of-the-art RL post-training systems.

cs.DC↗
arXiv2026

Toward Generalist Autonomous Research via Hypothesis-Tree Refinement

Scientific progress depends on a repeated loop of exploration, experimentation, and abstraction. Researchers test candidate directions, interpret the evidence, and carry the resulting lessons into later attempts. We study how an AI agent can run this loop autonomously over long horizons. We introduce Arbor, a general framework for autonomous research that combines a long-lived coordinator, short-lived executors, and Hypothesis Tree Refinement (HTR), a persistent tree that links hypotheses, artifacts, evidence, and distilled insights across time. The coordinator manages global research strategy over the tree, while executors implement and test individual hypotheses in isolated worktrees. As results return, Arbor updates the tree, propagates reusable lessons, refines the search frontier, and admits verified improvements. This design turns autonomous research from a sequence of local attempts into a cumulative process in which strategy, execution, and evidence are carried across time. We evaluate Arbor under Autonomous Optimization (AO), an operational setting where an agent improves an initial research artifact through iterative experimentation without step-level human supervision. Across six real research tasks in model training, harness engineering, and data synthesis, Arbor achieves the best held-out result on all six tasks, attaining more than 2.5x the average relative held-out gain of Codex and Claude Code under the same task interface and resource budget. On MLE-Bench Lite, Arbor reaches 86.36% Any Medal with GPT-5.5, the strongest result in our comparison.

cs.CL↗
arXiv2026

Understanding Generative Recommendation with Semantic IDs from a Model-scaling View

Recent advancements in generative models have allowed the emergence of a promising paradigm for recommender systems (RS), known as Generative Recommendation (GR), which tries to unify rich item semantics and collaborative filtering signals. One popular modern approach is to use semantic IDs (SIDs), which are discrete codes quantized from the embeddings of modality encoders (e.g., large language or vision models), to represent items in an autoregressive user interaction sequence modeling setup (henceforth, SID-based GR). While generative models in other domains exhibit well-established scaling laws, our work reveals that SID-based GR shows significant bottlenecks while scaling up the model. In particular, the performance of SID-based GR quickly saturates as we enlarge each component: the modality encoder, the quantization tokenizer, and the RS itself. In this work, we identify the limited capacity of SIDs to encode item semantic information as one of the fundamental bottlenecks. Motivated by this observation, as an initial effort to obtain GR models with better scaling behaviors, we revisit another GR paradigm that directly uses large language models (LLMs) as recommenders (henceforth, LLM-as-RS). Our experiments show that the LLM-as-RS paradigm has superior model scaling properties and achieves up to 20 percent improvement over the best achievable performance of SID-based GR through scaling. We also challenge the prevailing belief that LLMs struggle to capture collaborative filtering information, showing that their ability to model user-item interactions improves as LLMs scale up. Our analyses on both SID-based GR and LLMs across model sizes from 44M to 14B parameters underscore the intrinsic scaling limits of SID-based GR and position LLM-as-RS as a promising path toward foundation models for GR.

cs.AI↗
arXiv2026

Towards Verifiable Multimodal Deep Research: A Multi-Agent Harness for Interleaved Report Generation

Large Language Models (LLMs) have advanced autonomous agents from deep search, which retrieves concise factual answers, to deep research, which synthesizes scattered evidence into long-form reports. However, verifiable multimodal deep research remains challenging due to open-ended synthesis without deterministic ground truth and the need to interleave textual arguments with visual evidence. We propose Ptah, a multi-agent harness for interleaved report generation. Ptah orchestrates the lifecycle from user query to rendered web report through planning, research, and writing stages, where specialized agents construct visual-aware plans, collect claim-grounded evidence, maintain source-aligned images in a Visual Working Memory, and compose reports through declarative multimodal tool use. A verifier agent serves as the harness's acceptance function, enforcing factual grounding, citation fidelity, and cross-modal consistency throughout the workflow. We further introduce PtahEval, an evaluation protocol that augments existing benchmarks with image-level and presentation-level assessments. Experiments on deep research benchmarks show that Ptah produces more reliable, visually informative, and usable human-facing multimodal reports than strong baselines. Our code is released at https://github.com/SnowNation101/Ptah

cs.CL↗
arXiv2026

Self-supervised User Profile Generation for Personalization

Personalizing large language models (LLMs) has become a central challenge as LLMs are deployed across recommendation, search, dialogue, and content generation -- settings where the same query should yield different answers given different users. A promising route is to summarize each user's interaction history into a natural-language memory or profile and prepend it to the prompt to facilitate personalization. Existing methods learn such profile generators with explicit rewards derived from labeled downstream tasks, which are expensive and sparse as they require annotated supervision for every target task. In light of this challenge, we introduce Bidirectional User Modeling via Profiles (BUMP), a self-supervised framework that trains a profile generator without any downstream labels. Specifically, given a user's interaction history, we use GRPO to train an LLM to emit a free-form textual profile under a bidirectional in-batch ranking objective: a small LLM judge measures (i) how well the generated profile, used as a query, ranks the user's own held-out interactions above interactions from other users in the batch, and (ii) how well a held-out interaction, used as a query, ranks the user's own profile above profiles of other users. Both directions are scored with multi-positive NDCG and combined into a dense reward per rollout; other users in the batch supply free negatives, so every training example yields supervision from raw interaction logs alone. Evaluated on the LaMP benchmark, BUMP matches or outperforms closed-source APIs and prior methods relying on labeled rewards, while requiring no task label at training.

cs.CL↗
arXiv2026

Learning Temporal Causal Structure via Smooth Differentiable Optimization

Causal discovery with instantaneous effects in multivariate time series is challenging, as the instantaneous structure must be acyclic. Prior methods enforce this by either separating instantaneous and lagged estimation into multi-stage pipelines or imposing algebraic acyclicity constraints via complex augmented Lagrangian optimization, both of which incur high computational cost. In this work, we propose a different approach: we learn a differentiable permutation of variables using the Gumbel--Sinkhorn operator and triangularize the instantaneous coefficient matrix of a Structural Vector Autoregressive (SVAR) model in the learned order. This converts acyclicity from a hard constraint into a parameterization and keeps it valid throughout optimization. In doing so, our method enables unified, continuous optimization with gradient-based learning, leading to improved efficiency in time--series causal discovery. Across three real-world benchmarks, our method achieves the best overall performance compared with 12 baselines in both discovery accuracy and efficiency. On the large-scale benchmark, it further demonstrates strong scalability, achieving more than a 6x speedup over competing methods.

cs.LG↗
arXiv2026

Breaking Information Cocoons: A Hyperbolic Framework for Balancing Exploration and Exploitation in Recommender Systems

Modern recommender systems often create information cocoons, restricting users' exposure to diverse content. The central challenge is to balance content exploration and exploitation while allowing users to adjust their recommendation preferences. Ideally, this balance can be captured with a hierarchical representation, where depth search facilitates exploitation and breadth search enables exploration. However, existing approaches face two fundamental limitations: Euclidean methods struggle to capture hierarchical structures, while hyperbolic methods, despite their superior hierarchical modeling, lack semantic understanding of user and item profiles and fail to provide a principled mechanism for balancing exploration and exploitation. To address these challenges, we propose HERec, a hyperbolic framework that effectively balances exploration and exploitation in recommender systems. Our framework introduces two key innovations: (1) a semantic-enhanced hierarchical mechanism that aligns rich textual descriptions with collaborative information directly in hyperbolic space. Theoretical gradient analysis demonstrates that this alignment effectively leverages the underlying hyperbolic manifold structure, resulting in more accurate modeling of users and items; (2) an automatic hierarchical clustering mechanism by optimizing Dasgupta's cost, which discovers hierarchical structures without requiring predefined hyperparameters, enabling user-adjustable exploration-exploitation trade-offs. Extensive experiments demonstrate that HERec consistently outperforms both Euclidean and hyperbolic baselines, achieving up to 5.49% improvement in utility metrics and 11.39% increase in diversity metrics, effectively mitigating information cocoons.

cs.IR↗
arXiv2026

Turning Video Models into Generalist Robot Policies

Video generative models have emerged as a promising robotics backbone, capable of generating videos that depict the completion of complex tasks across embodiments and environments. Recent work proposes robot foundation models that jointly predict future observations and actions by finetuning video models with action-labeled data. In this paper, we test the limits of an alternative approach: leave the video planner as-is while training an embodiment-specific inverse dynamics model (IDM). This decoupling offers several natural benefits: the video planner remains embodiment-agnostic, different video models can be interchanged easily without re-training the IDM, and the IDM can be independently trained with readily available self-play data. We present a closed-loop, video-to-action policy that combines an action-free video world model with a carefully-designed IDM based on the robot embodiment Jacobian. We demonstrate that our IDM design is both data-efficient and scalable to high-dimensional action spaces. Our policy, which we coin the Video-to-Embodied Robot Action Model (VERA), achieves strong performance across simulated and real-world benchmarks, including zero-shot Panda arm manipulation and 16-DoF Allegro-hand dexterous cube re-orientation. The same video planner can be used across multiple embodiments by pairing it with different embodiment-specific IDMs. Our results show that decoupled video planning plus faithful video-to-action translation is a viable alternative route towards zero-shot, cross-embodiment, and generalizable robot control. More results are available on our project website: https://vera.csail.mit.edu.

cs.RO↗
arXiv2026

AgentFugue: Agent Scaling for Long-Horizon Tasks through Collective Reasoning

Recent progress on long-horizon agentic tasks has been driven largely by scaling up individual agents through stronger models, better tools, and more effective scaffolding. In contrast, much less is understood about scaling out: whether multiple peer agents, all targeting the same task, can become an additional source of capability without relying on explicit role specialization or workflow orchestration. We study this question and propose AgentFugue, a collective reasoning framework built around a shared reasoning hub. As peer agents explore the same task in parallel, the hub records concise notes on what each agent has established, attempted, or ruled out, and enables each agent to selectively access what other agents have discovered in a form useful for its current search. This design turns otherwise isolated trajectories into a connected ecology of reusable intermediate reasoning without requiring centralized planning. We instantiate the hub as a plug-in communication layer, trained with supervised fine-tuning and end-to-end reinforcement learning. Across the challenging long-horizon settings we study, AgentFugue improves over strong baselines. Our results suggest that collective reasoning can turn scaling out peer agent systems into a distinct source of capability gains, rather than merely a way of spending more compute.

cs.AI↗
arXiv2026

Sequential Data Augmentation for Generative Recommendation

Generative recommendation plays a crucial role in personalized systems, predicting users' future interactions from their historical behavior sequences. A critical yet underexplored factor in training these models is data augmentation, the process of constructing training data from user interaction histories. By shaping the training distribution, data augmentation directly and often substantially affects model generalization and performance. Nevertheless, in much of the existing work, this process is simplified, applied inconsistently, or treated as a minor design choice, without a systematic and principled understanding of its effects. Motivated by our empirical finding that different augmentation strategies can yield large performance disparities, we conduct an in-depth analysis of how they reshape training distributions and influence alignment with future targets and generalization to unseen inputs. To systematize this design space, we propose GenPAS, a generalized and principled framework that models augmentation as a stochastic sampling process over input-target pairs with three bias-controlled steps: sequence sampling, target sampling, and input sampling. This formulation unifies widely used strategies as special cases and enables flexible control of the resulting training distribution. Our extensive experiments on benchmark and industrial datasets demonstrate that GenPAS yields superior accuracy, data efficiency, and parameter efficiency compared to existing strategies, providing practical guidance for principled training data construction in generative recommendation. Our code is available at https://github.com/snap-research/GenPAS.

cs.LG↗