arXiv:2507.23033v2 Announce Type: replace Abstract: Spiking Neural Networks (SNNs) provide an energy-efficient computing paradigm for neural rendering, but existing spike-based Neural Radiance Field (NeRF) models usually use a fixed inference time step for all scenes. This fixed temporal budget is inefficient because NeRF follows a scene-specific training paradigm, and different scenes require different temporal capacities to preserve rendering quality. This paper proposes Pretraining-based Adaptive Time-step Adjustment (PATA), a scene-wise adaptive time-step training framework for spike-based NeRF. PATA parameterizes the target inference time step as a trainable variable and optimizes it through a two-stage training process. A hybrid input mode strengthens early time-step outputs, while full-step soft supervision, smoothed rendering loss, and temporal-budget loss jointly maintain rendering fidelity and reduce temporal computation. The learned target time step is shared by all ray samples within a scene, preserving the parallel rendering structure of NeRF. Experiments on INGP-NeRF and TensoRF backbones across Synthetic-NeRF, Mip-NeRF 360, and LLFF show that PATA consistently reduces inference cost while maintaining competitive rendering quality. PATA reduces the estimated inference energy by up to 57.57\% on INGP-NeRF and 68.90\% on TensoRF, demonstrating its effectiveness across different neural rendering representations.
Science Journals
arXiv:2512.22240v5 Announce Type: replace Abstract: Machine learning models are primarily judged by predictive performance, especially in applied genomics, where explanations are read as biological findings. In practice, reported gene panels are stabilised by averaging, ranking, or taking consensus over the many models a pipeline produces across cross-validation folds, tuning grids, and repeated runs. This raises an overlooked question: when two models achieve high accuracy, do they rely on the same internal logic, or reach the same outcome via different mechanisms? We introduce EvoXplain, a diagnostic framework that measures whether a pipeline's explanation is uniquely determined across repeated training and model selection. Rather than analysing a single trained model, EvoXplain treats explanations as samples drawn from the training and model selection pipeline itself, without aggregating predictions or constructing ensembles, and examines whether they form a single coherent explanatory basin or separate into multiple structured basins. We evaluate EvoXplain on a TCGA pan-cancer cohort and a within-cancer breast-cancer subtype task, using elastic-net Logistic Regression and gradient-boosted trees. Although all models reach about 98% accuracy, explanation structure differs across pipelines. Holding the data split fixed and varying only the regularisation strength, equally accurate Logistic Regression models separate into a few discrete, reproducible basins that recur across 100 data splits and carry distinct biological content, while the gradient-boosted pipeline converges to one basin. The same multiplicity appears within a single cancer subtype, from the ordinary tuning step alone. EvoXplain makes explanatory structure visible, revealing when an averaged consensus corresponds to no single trained model, and reframes interpretability as a property of the training pipeline rather than of any single model.
arXiv:2512.24780v2 Announce Type: replace Abstract: Neural networks trained with standard objectives exhibit behaviors characteristic of probabilistic inference: soft clustering, prototype specialization, and Bayesian uncertainty tracking. These phenomena appear across architectures -- in attention mechanisms, classification heads, and energy-based models -- yet existing explanations often rely on loose analogies to mixture models or post-hoc architectural interpretation. We provide a direct explanation. For any objective with log-sum-exp structure over distances or energies, the gradient with respect to each distance is exactly the negative posterior responsibility of the corresponding component: $\partial L / \partial d_j = -r_j$. The identity is algebraic, requiring only differentiability; it is a specialization of Fisher's identity, and its significance here is its address: standard neural objectives instantiate it without modification. The consequence is that gradient descent on such objectives performs generalized expectation-maximization implicitly, with responsibilities arising as gradients to be applied rather than auxiliary variables to be computed. This result unifies three regimes of learning: unsupervised mixture modeling, where responsibilities are fully latent; attention, where responsibilities are conditioned on queries; and cross-entropy classification, where supervision clamps responsibilities to targets. Our claims live at training time: the responsibility-weighted gradient dynamics recently documented in transformers follow from the objective's geometry. The in-context Bayesian computation that trained transformers perform at inference time is the endpoint of these dynamics, not their per-step content.
arXiv:2601.23087v5 Announce Type: replace Abstract: Learning long-horizon robotic manipulation requires jointly achieving expressive behavior modeling, real-time inference, and stable execution, which remains challenging for existing generative policies. Diffusion-based approaches offer strong modeling capacity but incur high inference latency, while flow matching enables fast, near-single-step generation yet often suffers from unstable execution when operating directly in the raw action space. We propose Continuous Latent Action Flow Policy (CoLA-Flow Policy), a trajectory-level imitation learning framework that performs flow matching in a continuous latent action space. By encoding action sequences into temporally coherent latent trajectories and learning an explicit latent-space flow, CoLA-Flow Policy decouples global motion structure from low-level control noise, enabling smooth and reliable long-horizon execution. The framework further integrates geometry-aware point cloud conditioning and execution-time multimodal modulation, using visual cues as a representative modality to enhance real-world robustness. Experiments in simulation and on real robots show that CoLA-Flow Policy achieves near-single-step inference, improves trajectory smoothness by up to 93.7% and task success by up to 25 percentage points over raw action-space flow baselines, while remaining significantly faster than diffusion-based policies.
arXiv:2604.26500v2 Announce Type: replace Abstract: LLMs and speech assistants are increasingly used for task-oriented interactions, yet their evaluation often relies on controlled scenarios that fail to capture the variability and complexity of real user requests. Drink ordering, for example, involves diverse named entities, drink types, sizes, customizations, and brand-specific terminology, as well as spontaneous speech phenomena such as hesitations and self-corrections. To address this gap, we introduce StarDrinks, a test set in English and Korean containing speech utterances features, transcriptions, and annotated slots. Our dataset supports speech-to-slots SLU, transcription-to-slots NLU, and speech-to-transcription ASR evaluation, providing a realistic benchmark for model robustness and generalization in a linguistically rich, real-world task.
arXiv:2607.01156v2 Announce Type: replace Abstract: We present the verification of a thermo--hydrologic--mechanical capability implemented within the PFLOTRAN framework, with emphasis on benchmark-based assessment of the THM implementation. The thermal--hydrologic (TH) equations for mass and energy balance are solved on control-volume blocks or Voronoi cells, while the quasi-static momentum balance is solved on an element-based dual mesh. The coupling is achieved using a strictly sequential, non-iterative fixed-stress split strategy in which the TH system is solved implicitly for pressure and temperature, followed by a mechanics update for the displacement unknowns. Several verification problems are set up against poroelastic and thermo-poroelastic benchmarks, demonstrating agreement with analytical or semi-analytical benchmark responses for pressure diffusion, the temperature field, and mechanical deformation. In addition, we propose a treatment for discontinuities (e.g., fractures) based on mapping between mechanical and flow degrees of freedom, and validate the approach by comparison to an analytical solution. This work establishes the basis for thermo-poroelastic coupling in PFLOTRAN and provides a solid modeling foundation for a range of applications (e.g., enhanced geothermal systems and other subsurface energy storage) involving coupled thermal--hydrologic--mechanical (THM) processes in geologic porous media.
arXiv:2607.04315v1 Announce Type: cross Abstract: This paper studies the problem of identifying the treatment that maximizes the expected natural direct potential outcome (NDPO), which captures the potential outcome of an intervention while excluding the pathway transmitted through a mediator that researchers may wish to remove from evaluation. We first establish population-level identification of the expected NDPO in a causal bandit setting using observable interventional distributions. We then develop a fixed-confidence best-arm identification (BAI) algorithm based on the Track-and-Stop (TaS) framework, employing a cutting-set method to solve the resulting semi-infinite optimization problem. The proposed algorithm achieves sample-efficient identification with a high-probability correctness guarantee. We prove that it satisfies $\delta$-correctness and asymptotic optimality. Finally, we validate the approach through empirical evaluations on a large-scale real-world advertising dataset (IPinYou).
arXiv:2607.04373v1 Announce Type: cross Abstract: Let $f_r(k)$ be the smallest $n$ such that every $r$-coloring of $\{1,2,\ldots,n\}$ has a monochromatic solution to the equation \[\frac{1}{x_1}+\frac{1}{x_2}+\cdots+\frac{1}{x_k}=\frac{1}{x_{k+1}}, \] where $x_1,x_2,\ldots,x_k$ are not necessarily distinct. In this paper, we prove that $f_r(2)\geq 4^r/2$ for all $r\geq1$, and $f_r(k)\geq(2^r-1)k^r$ for all $k\geq3$ and $r\geq1$. When $r=2$, we show that, if $k=3\cdot2^m$ for some positive integer $m$, then $f_2(k)=3k^2$; and if $k=p^m$ for some odd prime number $p$ and positive integer $m$, then $f_2(k)\geq3k^2+1$. We also provide new computational results for $f_2(k)$ and $f_3(k)$, as well as a generalization of our lower bounds for $f_2(k)$ to equations with general coefficients.
arXiv:2607.04936v1 Announce Type: cross Abstract: We investigate the use of twin optical fingerprints derived from correlated physical unclonable functions (PUFs), as a hardware-based platform for cryptographic key generation and distribution. Each fingerprint is associated with a random, yet reproducible speckle pattern, generated when coherent light is scattered by a disordered optical structure. We consider a pair of correlated optical PUFs, and study the conditions under which two honest parties can establish a common secret key, despite fabrication-induced variability and environmental noise. An explicit information-theoretic key-agreement protocol is developed, incorporating secure sketches, error reconciliation, and privacy amplification. We quantify information leakage due to public helper data, and derive lower bounds on the length of the final secret key. The analysis identifies the noise regimes in which secure key agreement is feasible, and examines the performance of both practical and near-capacity reconciliation schemes. Finally, we discuss how twin optical PUFs could be integrated into quantum key distribution (QKD) networks, as a mechanism for establishing an initial pre-shared secret key between two honest users, without relying on computational assumptions or trusted third parties.
arXiv:2607.04471v1 Announce Type: cross Abstract: Linear spatial filters (beamformers) enable robust, generalizable and interpretable speech enhancement with performance guarantees under ideal parameterization. Modern beamformers are often parameterized by deep neural networks, whose performance degrades in dynamic scenarios with multiple moving speakers of unknown directions. We propose a data-driven beamforming pipeline, which only requires an estimate of the target's initial direction. Building on a higher-order ambisonics representation, we show that neural temporal-spectral processing can be decoupled from linear spatial processing, and thereby achieve generalizable and array-agnostic enhancement. By incorporating autoregression into a frame-wise causal framework, we maintain consistent performance throughout fast speaker motion and long recordings. Evaluation on synthetic data demonstrates robust enhancement under challenging conditions with closely spaced and crossing speakers. Real-world recordings in a dynamic office meeting scenario complement these findings and show generalizability across varying ambisonics orders.
arXiv:2607.04497v1 Announce Type: cross Abstract: We present an open, reproducible branch-and-cut (B&C) algorithm for the capacitated profitable tour problem (CPTP) and its open s-t path variant, the capacity-constrained elementary shortest-path problem. The solver re-implements the formulation and cut families of Jepsen et al. (2014) on a fully open mixed-integer programming stack (HiGHS; Huangfu and Hall, 2018), and adds bound-based preprocessing, domain propagation, and reduced-cost variable fixing. We claim no new method; the contribution is twofold. First, an open, reproducible artifact: to our knowledge the first branch-and-cut for this problem class on a fully open stack, with the formulation, every separator, and all benchmark scripts released, so the results below can be rerun and the solver reused and extended as a baseline. Second, a component study on this common modern stack, benchmarked against a dynamic-programming/labelling reference, that decomposes which components pay off and where the running time goes. We find that the capacity-class cuts account for essentially the entire benefit (adding them to a connectivity-only baseline lifts the number of instances solved from 52 to 64 of 76 and shrinks the search tree more than tenfold), while comb and rounded generalized-large-multistar cuts, reduced-cost fixing, and bound-based propagation add nothing measurable. We also report a negative result: the shortest-path-incompatibility (SPI) cut, a variant of the node-precedence inequalities of Garc\'ia (2009), finds no violated inequality on any instance. The solver and all experiments are released as open, reproducible software (Spoorendonk, 2026).
arXiv:2602.14401v2 Announce Type: replace Abstract: Vision-Language Navigation VLN requires large-scale trajectory instruction data from private indoor environments, raising significant privacy concerns. Federated Learning FL mitigates this by keeping data on-device, but vanilla FL struggles under VLNs' extreme cross-client heterogeneity in environments and instruction styles, making a single global model suboptimal. This paper proposes pFedNavi, a structure-aware and dynamically adaptive personalized federated learning framework tailored for VLN. Our key idea is to personalize where it matters: pFedNavi adaptively identifies client-specific layers via layer-wise mixing coefficients, and performs fine-grained parameter fusion on the selected components (e.g., the encoder-decoder projection and environment-sensitive decoder layers) to balance global knowledge sharing with local specialization. We evaluate pFedNavi on two standard VLN benchmarks, R2R and RxR, using both ResNet and CLIP visual representations. Across all metrics, pFedNavi consistently outperforms the FedAvg-based VLN baseline, achieving up to 7.5% improvement in navigation success rate and up to 7.8% gain in trajectory fidelity, while converging 1.38x faster under non-IID conditions.
arXiv:2605.04701v2 Announce Type: replace Abstract: Preference restrictions have played a significant role in computational social choice. This paper studies a framework that connects preference restrictions with classical graph search paradigms. We model candidates as vertices of a graph and interpret the preference ordering of each voter as the outcome of traversing the graph according to a graph search. We focus on six fundamental paradigms: breadth-first search (BFS), depth-first search (DFS), breadth-first search (LexBFS), lexicographic depth-first (LexDFS), maximum cardinality search (MCS), and maximal neighborhood search (MNS). Within this framework, we study the problem of determining whether a given preference profile admits a graph support subject to structural restrictions, that is, whether there exists a graph such that each preference ordering can be generated by traversing the graph under the chosen paradigm. For all considered paradigms, we show that this problem is NP-hard when the graph support is required to have at most $k$ edges, where $k$ is a given integer. We further extend these hardness results to the case where the graph support is required to have maximum degree $k$. For DFS, we prove that recognizing whether a preference profile admits a tree support can be solved in polynomial time. Moreover, existing results imply polynomial-time solvability of the problem for all remaining graph traversals, except BFS and LexBFS, for which the complexity remains open.
arXiv:2509.14088v3 Announce Type: replace Abstract: Modern computing students often rely on both natural-language prompting and manual code editing to solve programming tasks. Yet we still lack a clear understanding of how these two modes are combined in practice, and how their usage varies with task complexity and student ability. In this paper, we investigate this through a large-scale study in an introductory programming course, collecting 13,305 interactions from 355 students during a three-day lab activity. Our analysis shows that students primarily use prompting to generate initial solutions, and then often enter short edit-run loops to refine their code following a failed execution. Student reflections confirm that prompting is helpful for structuring solutions, editing is effective for making targeted corrections, while both are useful for learning. We find that manual editing becomes more frequent as task complexity increases, but most edits remain concise, with many affecting a single line of code. Higher-performing students succeed with less reliance on editing and fewer overall interactions. These findings highlight the role of manual editing as a form of last-mile repair, complementing prompting in AI-assisted programming workflows.
arXiv:2607.05157v1 Announce Type: cross Abstract: Trace estimation is central in many lattice QCD computations, but the accuracy of the standard, stochastic Hutchinson method improves only with the square root of the sample size, making precise results expensive. We investigate two complementary variance reduction strategies. First, multigrid multilevel Monte Carlo uses a multigrid hierarchy to construct an unbiased multilevel estimator via recursive coarse grid corrections available from the multigrid hierarchy of the solver. Second, stochastic probing uses distance-$d$ graph colorings; we propose a torus based coloring that requires substantially fewer colors than hierarchical probing at the same distance. We test these approaches on two representative problems: the connected pseudoscalar correlator and disconnected fermion loops. For the connected pseudoscalar two-point function, the multilevel decomposition yields a variance reduction of up to $\mathcal{O}(10^5)$ at large time separations and translates into a clear cost reduction at fixed accuracy, thus confirming earlier results of arXiv:2412.06347. For the disconnected loops, in contrast, the multilevel decomposition provides only moderate gains, whereas probing combined with dilution delivers a substantial cost reduction that improves as the number of probing vectors is increased. Overall, the results highlight a pronounced complementarity: deflation schemes are most effective for observables dominated by long distance propagation, while probing is most effective for localized quantities.
arXiv:2607.05276v1 Announce Type: cross Abstract: Speaker embeddings, or x-vectors, are widely used to represent speaker identity and speaker-related attributes, but existing embedding extractors are typically descriptive rather than generative: they map an observed speech segment to an x-vector, which is then used for downstream applications. We introduce ProPS, Prompted Profile Synthesis, a framework for generating distributions of speaker embeddings conditioned on natural language prompts such as "a thirties male speaker with an Indian accent". ProPS converts human-written profile descriptions into sentence embeddings and uses a mixture density network trained on a large-scale dataset to predict a Gaussian mixture model in the x-vector space. The model is trained by maximizing the likelihood that real speaker embeddings match the requested profile, and its generated distributions are evaluated by negative log-likelihood on held-out x-vectors and by attribute classification accuracies on sampled synthetic x-vectors. Experiments show that ProPS produces profile-conditioned distributions and generates x-vectors that preserve requested speaker attributes such as age, gender, accent, and prosodic characteristics. This design enables controllable speaker-profile synthesis for speech generation systems like Text-To-Speech (TTS) or Voice Conversion (VC) while anchoring generated distributions in observed speaker-embedding structure.
arXiv:2607.04740v1 Announce Type: new Abstract: Spatial Transcriptomics (ST) measures gene expression while preserving spatial context, but its high cost and low throughput leave public datasets small. Inferring expression directly from widely available Hematoxylin and Eosin (H&E) stained histology offers a cost-effective alternative. However, existing approaches face several limitations: regression methods over-smooth toward the conditional mean, while generative methods are faithful but require slow multi-step inference; most methods treat genes as independent and equally important, ignoring inter-gene dependencies and heterogeneous gene informativeness; and most are tailored to a single resolution, either spot-level or cell-level. To address these issues, we propose DriftST, a unified framework for inferring spatially resolved gene expression from H&E images. DriftST builds on a Cellular Drifting generative model that learns a direct drift from a histology-conditioned source to the expression distribution, retaining generative expressiveness while enabling efficient one-step generation. To capture gene structure, we introduce the STransformer, which combines a co-expression attention module for inter-gene dependencies with a gene residual gate for differential gene importance. Operating on a generic gene-panel representation, DriftST applies directly to both spot-level and cell-level data in one framework, and extensive experiments across diverse tissues and platforms show that it achieves state-of-the-art performance at both resolutions.
arXiv:2607.01767v2 Announce Type: replace Abstract: Long-horizon language agents increasingly maintain executable world models in the form of planning graphs, where tool calls, validators, memory updates, recovery branches, and final answers are connected by typed dependencies. When a rollout fails, repairing the most visible error can leave the underlying error-amplification path intact, while replaying the full graph is expensive and difficult for long-context models to use reliably. We study world-model correction: selecting a compact subgraph of a failed planning graph whose repair stabilizes subsequent rollouts. We first instantiate a strong family of engineering correctors, including pointwise error scans, TopK and window selection, local graph expansion, cascade repair, and full-context LLM repair. We then propose WM-SAR, a spectral subgraph repair method that estimates node-edge amplification, greedily grows a connected repair region by marginal residual-spectral relief, and sends only this region to an LLM for root-cause repair. Theoretically, we connect residual spectral radius to rollout error and planning regret, motivating repair as stabilization rather than attribution alone. Across synthetic calling-tree graphs, benchmark-inspired agent topologies, and cross-model LLM repair experiments, WM-SAR achieves stronger long-horizon stabilization and root-cause recovery under compact token budgets, matching much larger repair contexts while exposing the LLM to a cleaner causal subgraph.
arXiv:2506.20116v2 Announce Type: replace-cross Abstract: Liquid marbles, droplets coated with a granular layer, are highly mobile as particles prevent capillary adhesion to the substrate. Yet their coating creates a static rolling friction, which we measure and model. Motion requires shear within the shell so that it is governed mainly by the grain density. This density controls yielding via a logistic function emerging as the particle network approaches percolation and increasing rapidly at close packing. More broadly, our system offers a simple platform for probing granular-raft mechanics and measuring their effective surface tension.
arXiv:2603.04795v2 Announce Type: replace Abstract: Medical image analysis depends on accurate segmentation and controllable synthesis, but both tasks face severe spatial imbalance: lesions occupy small regions against large backgrounds. We study adaptive spatial weighting as a task-level design principle and instantiate it in two adapters. LAW learns per-pixel loss weights for mask-conditioned diffusion by modulating a ratio prior with a feature-dependent delta map, with normalization, clamping, and Dice regularization for stability. ORDER improves lightweight segmentation by adding selective bidirectional skip attention with stage-wise confidence gating. On held-out diffusion test sets, LAW lowers FID from 158.13$\pm$0.15 to 108.43$\pm$0.71 on Polyps, from 144.13$\pm$0.31 to 89.51$\pm$0.96 on KiTS19, and from 139.22$\pm$0.38 to 112.58$\pm$0.68 on BRISC, while improving held-out mask-recovery Dice from 0.681$\pm$0.013 to 0.825$\pm$0.003 on Polyps. When the resulting images are added to nnUNet training, downstream Polyps mDice rises from 71.7$\pm$0.4 to 74.1$\pm$0.8. On the cleaned Polyps segmentation protocol, the reported ORDER configuration reaches 76.3$\pm$1.9 mDice and 67.2$\pm$2.0 mIoU at 42K parameters and 0.11 GFLOPs, versus 70.3$\pm$1.5 mDice and 59.9$\pm$1.7 mIoU for matched MK-UNet. On BRISC under the same training recipe, ORDER reaches 77.4$\pm$0.8 mDice and 68.1$\pm$0.7 mIoU. These results position adaptive spatial weighting as a practical design idea for both medical diffusion and efficient segmentation.
arXiv:2603.04836v2 Announce Type: replace Abstract: Modern e-commerce search is inherently multimodal: customers make purchase decisions by jointly considering product text and visual informations. However, most industrial retrieval and ranking systems primarily rely on textual information, underutilizing the rich visual signals available in product images. In this work, we study unified text-image fusion for two-tower retrieval models in the e-commerce domain. We demonstrate that domain-specific fine-tuning and two stage alignment between query with product text and image modalities are both crucial for effective multimodal retrieval. Building on these insights, we propose a noval modality fusion network to fuse image and text information and capture cross-modal complementary information. Experiments on large-scale e-commerce datasets validate the effectiveness of the proposed approach.
arXiv:2602.12262v4 Announce Type: replace Abstract: Diffusion large language models (DLLMs) have emerged as powerful generative models with the promise of fast text generation through parallel decoding. However, realizing this potential in practice remains challenging: reducing the number of decoding steps, typically causes a substantial degradation in output quality due to token factorization error. To alleviate this, we propose a self-distillation framework that trains a few-step student to match the generative trajectory of a full-step teacher. We theoretically and empirically show that trajectory-level supervision mitigates this factorization error, thereby enabling effective few-step decoding. We further incorporate Direct Discriminative Optimization (DDO), a reverse-KL objective that encourages mode-seeking toward the teacher's modes, yielding stronger performance on challenging reasoning tasks. Across reasoning and code-generation benchmarks, our method substantially narrows the gap between few-step and full-step decoding. The source code is available at https://github.com/Tyrion58/T3D.
arXiv:2602.12521v3 Announce Type: replace Abstract: Rail-optimized network fabrics have become the de facto datacenter scale-out fabric for large-scale ML training. However, the use of high-radix electrical switches to provide all-to-all connectivity in rails imposes substantial power and cost. We propose a rethinking of the rail abstraction by retaining its communication semantics, but realizing it using optical circuit switches. The key challenge is that optical switches support one-to-one connectivity at a time, limiting the fan-out of traffic in ML workloads using hybrid parallelisms. We overcome this through \emph{parallelism-driven rail reconfiguration}, which exploits the non-overlapping communication phases of different parallelism dimensions. This time-multiplexes a single set of physical ports across circuit configurations tailored to each phase within a training iteration. We design and implement Opus, a control plane that orchestrates this in-job reconfiguration of photonic rails at parallelism phase boundaries, and evaluate it on a physical OCS testbed, the Perlmutter supercomputer, and in simulation at up to 2,048 GPUs. Our results show that photonic rails can achieve over $23\times$ network power reduction and $4\times$ cost savings while incurring only modest training overhead at production-relevant OCS reconfiguration latencies.
arXiv:2601.21523v2 Announce Type: replace Abstract: To promote cooperation in Multi-Agent Reinforcement Learning, the reward signals of all agents can be aggregated together, forming global rewards that are commonly known as the fully cooperative setting. However, global rewards are usually noisy because they contain the contributions of all agents, which have to be resolved in the credit assignment process. On the other hand, using local reward benefits from faster learning due to the separation of agents' contributions, but can be suboptimal as agents myopically optimize their own reward while disregarding the global optimality. In this work, we propose a method that combines the merits of both approaches. By using a graph of interaction between agents, our method discerns the individual agent contribution in a more fine-grained manner than a global reward, while alleviating the cooperation problem with agents' local reward. We also introduce a practical approach for approximating such a graph. Our experiments demonstrate the flexibility of the approach, enabling improvements over the traditional local and global reward settings.
arXiv:2602.17284v2 Announce Type: replace Abstract: We consider the privacy amplification properties of a sampling scheme in which a user's data isused in $k$ steps chosen randomly and uniformly from a sequence (or set) of $t$ steps. This sampling scheme has been recently applied in the context of differentially private optimization (Chua et al., 2024a; Choquette-Choo et al., 2025) and communication-efficient high-dimensional private aggregation (Asi et al., 2026), where it was shown to have utility advantages over the standard Poisson sampling. Theoretical analyses of this sampling scheme (Feldman & Shenfeld, 2025; Dong et al., 2025) lead to bounds that are close to those of Poisson sampling, yet still have two significant shortcomings. First, in many practical settings, the resulting privacy parameters are not tight due to the approximation steps in the analysis. Second, the computed parameters are either the hockey stick or Renyi divergence, both of which introduce overheads when used in privacy loss accounting. In this work, we demonstrate that the privacy loss distribution (PLD) of random allocation applied to any differentially private algorithm can be computed efficiently. When applied to the Gaussian mechanism, our results demonstrate that the privacy-utility trade-off for random allocation is at least as good as that of Poisson subsampling. In particular, random allocation is better suited for training via DP-SGD. To support these computations, our work develops new tools for general privacy loss accounting based on a notion of PLD realization. This notion allows us to extend accurate privacy loss accounting to subsampling which previously required manual noise-mechanism-specific analysis.