arXiv subjects

Yifan Zhang

Publications and source records attributed to Yifan Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Agora: Git as Shared Memory for Collective AutoResearch

Research agents working in separate sessions need to know what others have tried and which results they can build on. Agora stores their contributions as an append-only directed acyclic graph (DAG) in Git. Each commit records a result, insight, hypothesis, verification, or report and links it to prior work. Searchable views show leading results, neglected branches, and verification status; diversity-aware recommendations suggest experiments beyond the current leaders. We report a run of nearly 12 days in which 13 language-model workers, with no assigned tasks or central planner, used Agora to solve a weight-transfer problem. Given 141 pretrained donor models and a frozen 119.6M-parameter attention--SSM hybrid whose dimensions match no donor, the workers had to initialize the target without training data or gradient updates. They published 1,703 contributions and reduced the development evaluator score from 3.39 to 1.899 bits per byte, closing 62% of the gap to a trained GPT-2 124M. The best method compresses donor next-token statistics into the target's embedding and output head, then adds short-range context through sparse edits to attention, feed-forward, and state-space blocks. Its 145-commit ancestry spans 15 accounts. Participants also posted 165 verifications of 95 targets, each by an account other than the target's author, with no reported failures. The run documents how agents reused and verified shared work. Measuring the effect on discovery per unit of compute requires a matched comparison.

cs.LG↗

arXiv2026

Completion-Aware Cross-Fidelity Offline-to-Online Reinforcement Learning for Multi-Line Bus Holding

Exploratory reinforcement learning (RL) on an operating bus fleet is impractical,while policies trained only from historical data cannot acquire new experience. Hybrid Offline-and-Online (H2O) RL combines fixed target replay with simulator interaction, but the inexpensive online simulator can differ from the target in transition and event-duration dynamics. We study this cross-fidelity problem for multi-line bus holding and address a failure mode in which lower generalized passenger time coexists with incomplete passenger journeys.

cs.AI↗

arXiv2026

Minimum-Excess $\{K_3,K_4\}$-Coverings of $K_{17}$, $K_{18}$, and $K_{19}$

We determine the minimum-excess coverings of $K_{17}$, $K_{18}$, and $K_{19}$ by 3-cliques and 4-cliques, minimizing first the number of repeated edge occurrences and then the number of blocks. The three secondary optima are $$ C^ξ(17,\{3,4\},2)=29,\qquad C^ξ(18,\{3,4\},2)=33,\qquad C^ξ(19,\{3,4\},2)=35. $$ For $K_{18}$ and $K_{19}$ the minimum excess is zero, so the optimal covers are decompositions, with block vectors $15K_3+18K_4$ and $13K_3+22K_4$, respectively. Their lower bounds follow directly from the known values $g^{(4)}(18)=33$ and $g^{(4)}(19)=35$ for pairwise balanced designs with maximum block size four. For $K_{17}$ the minimum excess is two and every optimal cover has block vector $12K_3+17K_4$. At this optimum the excess multigraph may be $P_3$ or $2K_2$, but not $K_2^2$; in fact a cover with excess $K_2^2$ requires at least $31$ blocks. The remaining order-$17$ nonexistence cases are excluded by computer-assisted exhaustive searches. Explicit constructions are given for all attainable optima. For completeness, an appendix also records the independent proof for orders $18$ and $19$ from the earlier version of this work, including its structural case analysis.

math.CO↗

arXiv2026

Matrix-game 2.0: An open-source, real-time, and streaming interactive world model

Recent advances in interactive video generations have demonstrated diffusion model's potential as world models by capturing complex physical dynamics and interactive behaviors. However, existing interactive world models depend on bidirectional attention and lengthy inference steps, severely limiting real-time performance. Consequently, they are hard to simulate real-world dynamics, where outcomes must update instantaneously based on historical context and current actions. To address this, we present Matrix-Game 2.0, an interactive world model generates long videos on-the-fly via few-step auto-regressive diffusion. Our framework consists of three key components: (1) A scalable data production pipeline for Unreal Engine and GTA5 environments to effectively produce massive amounts (about 1200 hours) of video data with diverse interaction annotations; (2) An action injection module that enables frame-level mouse and keyboard inputs as interactive conditions; (3) A few-step distillation based on the casual architecture for real-time and streaming video generation. Matrix Game 2.0 can generate high-quality minute-level videos across diverse scenes at an ultra-fast speed of 25 FPS. We open-source our model weights and codebase to advance research in interactive world modeling.

cs.CV↗

arXiv2026

Minimum-excess coverings of complete graphs by 3-, 4-, and 5-cliques

We study two-level optimal coverings of the complete graph $K_v$ by cliques of orders $3$, $4$, and $5$. The first level minimizes the excess, namely the number of repeated edge occurrences, and the second level minimizes the number of cliques among coverings with minimum excess. We first isolate the two-size problem for triangles and 4-cliques, where the main congruence and local-degree methods already appear and where nonzero excess is unavoidable for one residue class of $v$. This motivates the passage to quintuples. For $\{K_3,K_4,K_5\}$ coverings we determine the minimum excess for every $v$, use an edge-count reduction for the secondary optimization, obtain exact values in ten residue classes modulo $20$, and give bounds for the remaining classes.

math.CO↗

arXiv2026

Information Bottleneck-Guided Adaptive Hypergraph Transformer for Brain Disease Diagnosis

Exploring high-order correlations and long-range dependencies in brain networks holds significant value for both neuroscience research and clinical diagnosis. However, previous studies have lacked a unified integration of high-order and long-range dependency information in brain networks, and there is substantial redundancy behind various types of information. These issues limit their effectiveness in the diagnosis of brain diseases. To address this, we propose an Information Bottleneck-Guided Adaptive HyperGraph Transformer (IBAHGT). By incorporating the information bottleneck (IB) principle, this approach enables adaptive learning of high-order correlations and both short- and long-range dependencies within a unified framework for brain network analysis, achieving high-precision brain disease diagnosis. IBAHGT consists of three key components: an information bottleneck-guided adaptive hypergraph convolution, which introduces a novel hypergraph information bottleneck (HIB) principle to adaptively learn hypergraph message-passing weights between nodes and hyperedges, optimizes information flow and captures high-order information in brain networks that is maximally informative and minimally redundant (MIMR). The Transformer encoder captures global information within brain networks through the attention mechanism, specifically modeling short- and long-range dependencies. An information bottleneck-guided node-level adaptive fusion employs the IB principle to learn independent weights for each node, facilitating the fine-grained integration of high-order information and global information to obtain an efficient representation for downstream tasks. Extensive experiments demonstrate that the proposed method outperforms current state-of-the-art methods and can identify biomarkers for clinical applications.

cs.AI↗

arXiv2026

Temporal Forcing: 4D Representation Alignment for Vision-Language-Action Models

Long-horizon robotic manipulation requires vision-language-action (VLA) models to track scene states and their evolution beyond the current observation. However, simply conditioning policies on observation history does not guarantee that the history is effectively utilized: action supervision constrains what the policy should do, but only indirectly constrains what its history representations should retain. To resolve this, we present Temporal Forcing, a 4D representation alignment framework that explicitly supervises latent temporal states and their transitions. Specifically, we first introduce a history pathway that compresses past observations into compact latent tokens. We then align these tokens and current-frame features with geometric targets from a pretrained 4D foundation model, providing direct supervision at both the state and transition levels. The 4D foundation model and alignment heads are used only for training-time supervision. Temporal Forcing improves average success from 96.6% to 98.8% on LIBERO, with the largest gain on LIBERO-Long (93.8% to 97.2%), and from 53.5% to 62.8% across twelve RoboTwin 2.0 tasks. Furthermore, Temporal Forcing increases full-task success from 20.0% to 43.3% on a physical multi-stage hidden-placement task. Controlled experiments show that 4D representation alignment is crucial for making observation history beneficial to the model. Code will be publicly available.

cs.RO↗

arXiv2026

SymbolicArena: A Unified Infrastructure for Benchmark Distillation and Dynamic Evaluation in Symbolic Regression

Symbolic regression (SR) seeks concise and interpretable mathematical expressions from data for scientific equation discovery. Existing SR benchmarks face a tradeoff between evaluation cost and benchmark validity. Repeated evaluation of large task pools is expensive, and compact benchmarks lack systematic evidence of preserved task diversity and algorithm discriminability. SymbolicArena provides a unified infrastructure for benchmark distillation and dynamic evaluation. The framework standardizes 664 heterogeneous tasks with executable ground truth expressions and distills the Full Task Set into Core50, a validated benchmark of 50 tasks. The distillation process preserves task coverage and algorithm discrimination under explicit balance constraints. SymbolicArena applies a unified execution protocol to heterogeneous SR algorithms and produces comparable outputs and search trajectories. Multi Axis Evaluation characterizes numerical quality, symbolic quality, and search behavior. Core50 reduces evaluation workload by 92.5% and maintains agreement with Full Task Set evaluations. Experiments show that SymbolicArena achieves 72.6% to 86.7% lower approximation error than alternative selectors, further supporting its fidelity to the Full Task Set. Evaluation reveals a substantial gap between numerical fitting and symbolic recovery across current SR methods, suggesting that reliable equation recovery remains an open challenge.

cs.LG↗

arXiv2026

Borrowing from anything: A generalizable framework for reference-guided instance editing

Reference-guided instance editing is fundamentally limited by semantic entanglement, where a reference's intrinsic appearance is intertwined with its extrinsic attributes. The key challenge lies in disentangling what information should be borrowed from the reference, and determining how to apply it appropriately to the target. To tackle this challenge, we propose GENIE, a Generalizable Instance Editing framework capable of achieving explicit disentanglement. GENIE first corrects spatial misalignments with a Spatial Alignment Module (SAM). Then, an Adaptive Residual Scaling Module (ARSM) learns what to borrow by amplifying salient intrinsic cues while suppressing extrinsic attributes, while a Progressive Attention Fusion (PAF) mechanism learns how to render this appearance onto the target, preserving its structure. Extensive experiments on the challenging AnyInsertion dataset demonstrate that GENIE achieves state-of-the-art fidelity and robustness, setting a new standard for disentanglement-based instance editing.

cs.CV↗

arXiv2026

Not Every Term Adds New Structure: Sobolev Novelty for Symbolic Regression

Symbolic regression (SR) aims to discover compact and meaningful mathematical equations from data, but searching the vast combinatorial space of symbolic structures remains challenging. Existing methods typically guide this process using expression-level objectives, such as fitting error, which assess a candidate equation as a whole but provide little information about whether an individual term contributes genuinely new structure or is largely redundant with the rest of the expression. We introduce \textbf{Sobolev Novelty}, a term-level measure of structural independence for symbolic equations. For each term, we construct an empirical Sobolev signature from its function values and exact derivatives over the observed inputs, and quantify how much of this behavior cannot be reconstructed by the remaining terms. We further derive a theory-calibrated threshold, yielding a principled and tuning-free criterion for identifying structurally novel terms. Using this threshold, 92.6\% of terms in benchmark ground-truth equations exhibit sufficient structural novelty, compared with only 38.3\% on average for expressions produced by 15 SR methods, revealing a substantial gap between scientific equations and current SR solutions. As a lightweight plug-in, Sobolev Novelty can be incorporated into diverse SR paradigms to support term pruning, search guidance, LLM feedback, and data selection, yielding consistent performance gains and demonstrating broad applicability.

cs.LG↗

arXiv2026

Refreshing Less, Selecting Better: Reusing Stale Gradient Features for Efficient Influence-Based Data Selection

Gradient-based data selection methods such as LESS score each candidate by the alignment between its gradient and a target validation gradient, and recomputing per-example gradient features at every new checkpoint dominates their cost. Across three selection seeds, two model families, two candidate pools, and two target tasks, features cached at a post-warmup checkpoint and paired with fresh validation gradients preserve the ranking 40 optimizer steps later with Spearman correlation from 0.952 to 0.991, while the top-10% subset they induce misses 10 to 22% of the examples that full recomputation selects. We therefore propose Cached Diverse Influence Selection (CDIS), which recomputes gradient features for the top-ranked fraction $p$ of candidates under the stale scores, fits an affine calibration on the recomputed examples, and selects the final subset under source and length quotas. A refresh fraction at or above the selection fraction recovers the exact top-$k$ subset whenever the calibrated stale scores have bounded error, and the budget curves follow this rule: at $p=0.3$ the recovered top-$k$ subset coincides with full recomputation in every setting, allocating the same budget per stratum recovers the stratified subset at 0.92 to 1.00, and gradient-stage wall-clock drops 3.5 to 3.6 times. Iterating the cache over four checkpoints keeps top-$k$ overlap at 0.98 or higher at 1.9 gradient features per example against 4 for full recomputation, and stale-to-recomputed agreement on the refreshed examples provides a free check for unsafe reuse. Downstream, unconstrained top-$k$ selection collapses to a single data source and scores 13 points below random selection on GSM8K. CDIS scores 12 points above random selection with paired confidence intervals that exclude zero and trails full recomputation by 4.3 points, one training-run standard deviation, at 3.4 times lower selection cost.

cs.LG↗

arXiv2026

Deep Delta Learning

Transformer residual streams are updated by addition. A sufficiently expressive residual block can represent content replacement, but the residual update itself has no operation that reads, compares, and replaces content. We introduce Deep Delta Learning (DDL), which applies the delta rule over network depth. Each layer reads the residual state along a learned direction, compares the readout with a learned target, and writes a gated rank-1 correction back along the same direction. A closed gate gives the identity map, and a unit gate overwrites the selected readout with the target. DDL works with the usual vector state or with an expanded state that stores several value channels, while attention and MLP blocks keep the original model width. We pretrain decoder-only models with approximately GPT-2 small and medium sizes on the FineWeb-Edu dataset. At both scales, every DDL variant has lower validation loss and higher average one-shot accuracy than the additive baseline, and the best expanded variants raise that average by 0.91 and 1.18 points.

cs.LG↗

arXiv2026

FlashSampling: Fast and Memory-Efficient Exact Sampling

Sampling from a categorical distribution is mathematically simple, but in large-vocabulary decoding, it often triggers extra memory traffic and extra kernels after the LM head. We present FlashSampling, an exact sampling primitive that fuses sampling into the LM-head matmul and never materializes the logits tensor in HBM. The method is simple: compute logits tile-by-tile on chip, add Gumbel noise, keep only one maximizer per row and per vocabulary tile, and finish with a small reduction over tiles. In tensor-parallel decoding, FlashSampling replaces the all-gather of logits with streaming peer-to-peer writes: This overlaps GPU-to-GPU communication with computation and HBM loads across up to 8 GPUs, with near-ideal scaling at large batch sizes. Our kernel is exact because $\text{argmax}$ decomposes over partitions; grouped variants for online and tensor-parallel settings are exact by hierarchical factorization of the categorical distribution. FlashSampling demonstrates kernel-level speedups on decode workloads across 4 different datacenter GPUs (H100, H200, B200, B300), and in end-to-end vLLM experiments, it reduces time per output token by up to $10%$ on the models we test. These results show that exact sampling, with no approximation, can be integrated into the matmul itself, consolidating the bandwidth-bound sampling step in an efficient epilogue.

cs.LG↗

arXiv2026

SPO: Discovering Adaptive Large Neighborhood Search Operators via Stackelberg Program Optimization

Large neighborhood search (LNS) relies critically on destroy and repair operators, whose effectiveness depends on both adaptation to the evolving LNS state and interaction between the two roles. We introduce Stackelberg Program Optimization (SPO), an LLM-based framework for discovering adaptive executable destroy-repair programs. SPO conditions operator decisions on a compact LNS state, allowing state-dependent behavior to emerge through program discovery, and organizes destroy-repair discovery as a Stackelberg interaction over program space that reflects their asymmetric dependency. Role-specific credits evaluate destroy programs as leaders and repair programs as conditional follower responses, guiding a coupled optimization process that combines LLM generator learning with population-based evolutionary search over programs. Experiments on the traveling salesperson problem and capacitated vehicle routing problem show that SPO outperforms strong baselines across a broad range of settings and generalizes beyond the discovery scale to larger instances and benchmark sets. Behavioral analyses further demonstrate state-dependent operator behavior and coupled destroy-repair improvement during discovery.

cs.AI↗

arXiv2026

More Sensors Only One Field: Rethinking Continual Spatio-Temporal Forecasting

Continual spatio-temporal forecasting supports traffic management and environmental monitoring under evolving dynamics and expanding sensor networks. However, conventional graph-based continual learning methods tie forecasting representations to the current sensor layout, so sensor expansion can alter the representation of learned spatial relationships. Our key insight is that sensor expansion changes the evidence available about a process without necessarily changing the dynamics to be learned. We propose STFO (Spatio-Temporal Field Operator), which parameterizes forecasting knowledge as a shared field-evolution operator and handles changing sensor layouts through observation and query interfaces. Normalized coordinate-based aggregation lifts irregular sensor histories onto a fixed latent grid, enabling reuse of learned spatial maps across observation sets without sensor-specific parameters. To accommodate process drift, a spectral descriptor summarizes variation across spatial scales and conditions Fourier propagation and attention to adapt operator responses to the current spatial regime. Coordinate-based decoding queries the evolved field at sensor locations and combines spatial corrections with local-history predictions. Experiments on PEMS-Stream, CA-Stream, and AIR-Stream demonstrate state-of-the-art average forecasting performance. STFO-Large reduces average MAE over DOL by 8.4% on PEMS-Stream and 4.7% on CA-Stream. Our code is available at https://github.com/Xielewei/Spatio-Temporal-Field-Operator.

cs.LG↗

arXiv2026

Depth-Based Local Recentering of Maxmin Landmarks for Lazy Witness Persistence

We study a local post-processing rule for maxmin (farthest-point) landmark selection in lazy witness persistence. Maxmin landmarks induce nearest-landmark cells. Within each cell, the landmark is moved toward a data point of maximum empirical halfspace depth, with smaller moves in sparsely populated cells, and is then projected back to a data point in the same cell. The resulting support-weighted rule preserves one landmark per original maxmin cell. We prove a convex-core guarantee for the depth target and explicit worst-case cover bounds for the recentered landmarks. Experiments on planar synthetic loops and 120 MPEG-7 (Moving Picture Experts Group 7) silhouettes separate average landmark fit from extremal coverage. On the synthetic benchmark, recentering reduces mean nearest-landmark distance in all 900 matched trials; after aggregation over landmark budgets, the mean change is $-0.0138$ across 300 dataset/noise/seed blocks, while maximum cover increases by $0.0067$ on average and a thresholded first homology ($H_1$) count is unchanged. On MPEG-7, the image-level mean distance decreases for all 120 silhouettes, while maximum cover increases for 118. Thus the correction consistently improves average landmark representativeness but does not improve worst-case coverage; on the controlled synthetic benchmark, this trade-off occurs without a detectable change in the coarse $H_1$ summary.

cs.CG↗

arXiv2026

Intrinsic {Right} $q$-Radial Vector Derivatives and Localized Fischer Decompositions on Radial Algebras

We define intrinsic right $q$-radial vector derivatives on radial algebras of abstract vector variables. For a finite parameter set $Y$, a scalar Jackson calculus on $x^2$ and the mixed anticommutators $\{x,y_i\}$ is combined with the exterior decomposition relative to $x$. Relabelling covariance and compatibility under inclusions of parameter sets give a direct-limit operator on arbitrary radial algebras. {Under the dimension specialization $Q=q^M$, its classical limit agrees in every $M$-dimensional Clifford realization with the standard right Clifford derivative; for an exterior blade one obtains $[x\wedge y_I]\partial_x=(-1)^{|I|}(M-|I|)y_I$.} Two Fischer-type constructions are considered. The anticommutator with exterior creation is triangular and admits an explicit Green inverse after localization at its diagonal factors. {For right $q$-monogenic elements, the derivative is paired with right multiplication $R_xF=Fx$. The homogeneous operators $\partial^{Y,\mathrm R}_{x,q}R_x$ have generically nonzero determinants, which yields determinant-localized Fischer decompositions and explicit projectors.} The one- and two-vector determinants are computed explicitly. {After $Q=q^M$, the two-vector determinant is nonzero for every $0<q<1$ and every positive integer $M$. For arbitrary finite support, a support filtration factors the determinant into exact-support terms; in degree zero, support rank $p$ contributes the factor $[m-p]_q+p$.}

math.CV↗

arXiv2026

Right q-vector calculus at integral superdimension: localized decompositions and resonance

We specialize the intrinsic right $q$-vector derivative on radial algebras to integral superdimension. The formal dimension is encoded by an independent coefficient $Q$, and formal radial superspace of superdimension $M=m-2n$ is obtained by the coefficient specialization $Q\mapsto q^M$. This gives a rigorous universal calculus and, on finite blocks containing at most $m$ abstract vectors, a faithful coordinate realization on $\mathbb R^{m|2n}$. The localized exterior result is formulated as a Green decomposition by complementary projector images. Whenever the specialized finite determinant is nonzero, full left multiplication yields a determinant-localized right-monogenic Fischer decomposition. Beyond this base-change theory, we determine the exceptional one-vector calculus completely: for $M=-2\ell$ there is one additional singular monomial and one missing image monomial, whereas all other integral superdimensions give a surjective derivative with constants as its kernel. We then prove that the degree-zero Fischer operator is exactly diagonal on exterior blades, obtain its determinant explicitly, classify all support-resonance values in $0<q<1$, and give an exact kernel-rank formula as a sum of support multiplicities. On a block with $N$ auxiliary vectors, an even support rank $p$ has pure multiplicity $\binom Np$ on the support truncation. At an odd-support root, every lower odd factor is nonzero and any simultaneous lower resonance is unique, even, and characterized by one strictly monotone scalar equation. These results distinguish persistent nonpositive-even-superdimension defects from isolated support-dependent $q$-resonances. An appendix records that constant scalar projection of two independent orthogonal right $q$-vector derivatives does not descend to the Hermitian quotient.

math.CV↗