In briefCollective Wisdom from Deliberating Jurors: Does Discussion Improve Group Accuracy?
SummaryThe paper presents a formal model of deliberation in which agents probabilistically acquire evidence and selectively disclose it to peers to induce opinion change. Using a jury theorem and simulations, the authors examine whether deliberation improves collective accuracy. They identify conditions under which wisdom of crowds emerges after a few deliberation steps. The effect is shown to be sensitive to the evidence acquisition mechanism and the diversity of the initial evidence distribution.
In briefA decremental algorithm for detecting Braess paradox risk in changing traffic networks
SummaryThe paper studies the Braess paradox, where removing road links lowers travel latency at Wardrop equilibrium. It provides a decremental algorithm for checking this network vulnerability in dynamically evolving networks, with linear amortized cost per edge removal. The authors also improve a static vulnerability-checking algorithm from O(n m^2) to O(m^2).
In briefDIALER: Improving Rare-Class Accuracy in Retraining-Free Edge Video Analytics
SummaryLightweight models used for edge video analytics lose accuracy as live video streams shift in distribution, and continuous retraining strains limited edge compute. The paper finds that vision foundation models offer high average accuracy without retraining and with substantial compute savings, but often misclassify critical rare classes as similar common ones. The authors present DIALER, which uses spare GPU cycles to run pre-built multi-stage corrections for common rare-to-common confusion pairs. On four real-world driving datasets, it improves rare-class accuracy by up to 14.0% without interfering with real-time multi-stream inference.
In briefFrom Last-Week Add-On to Day-One Practice: Three Years of Teaching MPI Performance Analysis with EduMPI Suite
SummaryThis paper reports three years of using EduMPI Suite in HPC education, which automates cluster execution, measurement, and near-real-time analysis and visualizes MPI communication from the first program. The authors trace how performance analysis moved from a last-week add-on to a recurring part of MPI development. In the 2025 study, 29 to 33 of 33 students rated the tool higher than conventional tools, and all recommended introducing it early in the course.
In briefStable matching algorithm for complementary preferences with misaligned interests
SummaryThis paper studies many-to-one two-sided stable matching where firms and workers rank the same allocations differently and preferences are complementary. The authors show a stable matching exists for every instance when each firm has an anchor worker and a Transitive Alignment condition holds. They propose CDAR, a Deferred Acceptance-style algorithm allowing reproposals, and prove it converges in finite time and outputs a stable matching. Under strict preferences, its output is Pareto optimal among stable matchings.
Tatsuya Iwase, Baharak Rastegari, Sebastian Stein et al.
In briefSpeeding Up Data Analytics on Heterogeneous Supercloud Systems
SummaryThis paper presents computational models and performance estimation techniques for accelerating Dense Cholesky and SVD-based data analytics on heterogeneous Supercloud systems. The ReDSEa tool-chain automates mapping, load balancing, scheduling, parallelism, and computation overlap. On a system with a Huawei Kunpeng 920 ARM CPU and an Ascend 910 AI accelerator, the LLVM-based compiler tool-chain reports speedups of up to 17x for CH and 88x for SVD over fully optimized 48-core CPU implementations.
In briefShill-proof and non-shill-proof equilibria coexisting in a public auction
SummaryThe paper gives an explicit single-item auction example in the Komo, Kominers, and Roughgarden model where both a shill-proof and a non-shill-proof equilibrium exist. The example uses three bidder positions, publicly observed actions, positive winning prices, and a strictly regular i.i.d. value distribution. The authors verify both equilibria and compute their expected revenues.
In briefServe now or improve later? Scheduling self-evolution for online agents on shared GPUs
SummaryThis paper studies how online agents can use idle GPU capacity to build reusable tools or model states while still serving current requests. It proposes LearnSched, a state-aware scheduler that weighs the reuse window of a capability against the timely return of compute, choosing among progressing, checkpointing, or waiting. In 1920 finite-model scenarios based on A100 measurements, waiting for longer execution windows improved net value when cold recovery was expensive, while a strong window policy matched it under warm recovery.
In briefMMS allocation of indivisible items when items are few
SummaryThe paper studies allocating m indivisible items among n agents so that each agent receives at least her maximin share (MMS). It defines μ(n) as the largest m for which every instance with n agents and m items admits an MMS-satisfying allocation. For general monotone valuations, μ(n) = n + Θ(log n), holding for both goods and chores. For arbitrary (non-monotone) valuations, μ(n) = log n + O(1).
In briefFast, near-exact way to compute tournament placement odds for thousands of players
SummaryThe Independent Chip Model converts poker tournament chip stacks into finishing-place probabilities and prize equities. This paper presents DE-ICM, a deterministic algorithm that computes all players' placement probabilities for all paid places in O(M n²) time. On exactly solvable instances with up to 4,000 players, the worst relative equity error was 4.0×10⁻¹⁴ and the worst absolute placement error was 1.1×10⁻¹⁴. The full placement matrix for 1,000 players takes 0.2 seconds on one CPU core.
WhyExtends exact ICM placement computation to thousands of players via double-exponential quadrature and stable deconvolution, validated on instances up to 4,000 players with errors near 10^-14.
In briefHow hard is it to compute Nash equilibria in Colonel Blotto games?
SummaryThe paper studies how hard it is to compute Nash equilibria in discrete Colonel Blotto games. It shows that approximate equilibria are PPAD-hard even with only two battlefields, with PPAD-completeness for the hardness results. It also identifies a tractable case when the social payoff has constant rank, and proves hardness with only five players under equal splitting among highest bidders.
WhyResolves open complexity questions for Colonel Blotto equilibria, proving PPAD-hardness and tractable frontiers with rigorous proofs, affecting a niche game-theory subfield.
In briefSignaling dimension of cross-polytopes: how many classical resources are needed to mimic them
SummaryThis paper studies the signaling dimension of cross-polytopes in the general probabilistic model of information theory. The signaling dimension is the smallest classical communication complexity needed to perfectly reproduce a system's input-output correlations. It was previously known to be 3 for three, four, and five dimensional cross-polytopes. The authors give logarithmic lower and upper bounds for higher dimensions, determining the value up to an additive gap of 2. The lower bound shows the signaling dimension grows without bound, unlike balls and hypercubes.
In briefCodes that avoid conflicts, and near-primitive roots: a math paper on a number tied to code design.
SummaryThe paper studies a quantity from the theory of conflict-avoiding codes, namely the multiplicative order of 4 modulo d, where d ranges over the positive divisors of an odd integer n. Using methods from research on Artin's primitive root conjecture, the authors establish results on how this quantity is distributed as n runs over odd primes and over odd integers. They also give other interpretations, one of which connects it to the factorization of X^n-1 into irreducibles over the field F_2.
C. G. Karthick Babu, Pieter Moree, Sunil Kumar Pasupulati et al.
In briefPolynomials that vanish most often on a grid, with limited monomial support
SummaryThe paper studies extremal polynomials over a finite Cartesian set S = X×Y whose monomial support lies in a given decreasing set, choosing those with the most zeros in S. Coordinate factorizations give canonical extremal polynomials, and the authors characterize when noncanonical ones exist. This condition is stricter than the classical Prouhet–Tarry–Escott problem. The authors also count quadratic noncanonical extremal polynomials and apply the results to minimum-weight codewords in coding theory.
Dipak K. Bhunia, Eduardo Camps-Moreno, Ignacio Garc\'Ia-Marco et al.
In briefOn whether Omura's and Dueck–Körner's strong converse exponents match above capacity
SummaryThe paper studies a lower bound on the probability of correct decoding above channel capacity, Omura's sphere-packing-type exponent, and its refinement by Dueck and Körner. It notes that a claimed equality of the two exponents at all rates contains an error. The authors show that beyond a threshold rate the strong converse exponent is a line of unit slope set by the order-infinity Rényi capacity, and that the two exponents coincide up to a higher second threshold and differ strictly beyond it. Examples include modulo-additive channels, with no gap, and the binary erasure channel, where thresholds and the gap are given in closed form.
In briefMOLT: A fine-grained GPU memory sharing system for LLM serving and opportunistic fine-tuning
SummaryThe paper notes that GPU memory sits idle inside large language model serving replicas, and that existing approaches either keep tuning memory resident or reclaim it per whole training sample, discarding the running tuning step. The authors present MOLT, which lets inference reclaim memory of individual activations saved for a tuning step's backward pass, so the step continues and recomputes them. On four 24B–70B deployments across H100 SXM and B200 GPUs, MOLT reportedly keeps inference SLO attainment at or above 99.7% and completes 1.9–3.3x the tuning work of discard-based memory sharing.
In briefCodes that fix substring edits and optimal burst-deletion codes
SummaryThis paper studies codes that can correct a k-substring edit, which deletes a substring of length at most k and inserts a sequence of length at most k at the same position. The authors introduce a new localization method and construct q-ary codes with log n + 8 log log n + o(log log n) bits of redundancy, improving on prior binary results. When deleted and inserted lengths differ, they achieve redundancy log n + O(1), optimal up to an additive constant, which also yields optimal burst-deletion and localized-deletion codes.
WhyA new localization method achieves optimal log n+O(1) redundancy for q-ary substring edit codes and burst-deletion models, with rigorous proofs.
In briefPayment design when bidders disagree on beliefs: rent placement, bid resolution, and auction choice
SummaryThe paper studies how a seller can change not only the amount of information rent paid to bidders but also the states in which it is delivered, when bidders misperceive rivals. Payment design is formulated as a linear program over nonnegative bidder rents, with analysis under finite bidding interfaces and nonnegative-payment restrictions. A calibration using 1,000 eBay proof-set auction records shows that first-price auctions can outperform exactly truthful, no-subsidy rules, or the reverse, depending on how bid bands are partitioned.
In briefFactoring stable-matching instances into prime blocks, with an exact count of decomposable cases
SummaryThe paper shows that every balanced stable marriage instance with strict complete preferences has a unique finest partition into prime blocks. This one partition simultaneously factors the execution digraph as a Cartesian product, the proposal-prefix antimatroid as a direct sum, and the stable-matching lattice as a direct product. It gives an exact census for four agents per side, finding 88,478,208 decomposable profiles out of 110,075,314,176. The blocks are characterised via square components of the mutual-rank filtration, so they can be computed in polynomial time.
In briefHandling missing data in performance portability studies
SummaryThis paper addresses missing data in high-performance computing (HPC) performance portability studies. Gaps in performance efficiency datasets arise from incomplete benchmarking, hardware constraints, or implementation gaps, and many existing performance portability metrics cannot handle missing inputs. The authors evaluate a range of imputation methods and algorithms and apply them to real-world HPC datasets. They compare trade-offs between accuracy, robustness, and computational cost, and offer practical recommendations.
In briefSender-Receiver Games with Payment Commitments: Players pre-commit to payments tied to verified outcomes.
SummaryThis paper studies sender-receiver games where players commit in advance to payments based on the later-verified state and the receiver's action, with payments either transferred or burned. The authors analyze equilibrium guarantees, welfare, and computational complexity across different commitment structures. For binary actions they characterize equilibrium payoffs and identify which optimization problems are solvable efficiently versus NP-hard.
WhyIntroduces payment-commitment sender-receiver games, characterizing equilibrium guarantees and welfare with tight complexity results, including linear-time and NP-hardness findings, a notable theoreti
In briefInterpretable nonlinear feedback codes for AWGN channels using posterior means
SummaryThis paper proposes a posterior-mean feedback coding framework for communication over AWGN channels with feedback. It builds nonlinear feedback codes via posterior-mean refinement and uses MAP decoding at the receiver, under both noiseless passive and noisy active feedback. The scheme is analytically described and needs only a few learned power-allocation parameters, which the authors describe as interpretable. Numerical results show strong finite-blocklength performance, outperforming several analytical and learned baselines.
Yingyao Zhou, Natasha Devroye, Milos Zefran et al.
In briefOne-shot beam tracking for stacked metasurface-assisted satellite broadcasting
SummaryThis paper proposes a physics-informed one-shot beam tracking method for low-Earth-orbit satellite links assisted by a multilayer stacked intelligent metasurface. It predicts signal-quality changes from orbital motion and maps them to the metasurface phase settings through an analytical Jacobian, requiring only a single regularized correction. Numerical results show it keeps about 92% to 94% of the rate of iterative reoptimization while running hundreds of times faster.
Chandan Kumar Sheemar, Giovanni Iacovelli, Sourabh Solanki et al.
In briefLarge-scale analysis of IO500 benchmark results for storage performance
SummaryThis paper analyzes 294 post-reset IO500 submissions from 116 sites between 2019 and 2025 to study storage performance. It examines how individual benchmark phases relate to composite rankings. Phases measuring similar I/O behavior are strongly correlated, while bandwidth and metadata performance show weaker, scale-dependent relationships. Temporal and log-level analysis reveals information not visible in aggregate scores.
Aasish Kumar Sharma, Anila Ghazanfar, Sepehr Mahmoodianhamedani et al.
In briefOptimistic reliable broadcast with three-step latency
SummaryThis paper studies latency tradeoffs in signature-free asynchronous reliable broadcast that allow optimistic two-step delivery. The authors present two constructions that improve optimistic, honest-broadcaster, and totality delay bounds within specific ranges of parties and faults. They also prove a matching lower bound showing that, at maximal optimism, totality delay is at least three steps. The results indicate that maximal optimism costs one extra totality step compared with Bracha's protocol without increasing good-case latency.
WhyGives new signature-free reliable broadcast constructions with optimistic two-step delivery and a matching lower bound proof, a clear theoretical improvement.
In briefModulation and Integration in a Local Processor Receiving Two Input Sources
SummaryThis paper uses partial information decomposition to analyse how a local processor that receives two integrated input streams transmits information. The authors revise earlier conditions for contextual modulation and add a condition for signal integration, computing exact information-theoretic results with the Ibroja method. They test these conditions across several activation functions and apply them to noisy simulation data from a model layer 5b pyramidal cell, reporting evidence of basal-apical modulation and integration.
In briefCertified Splitting: Checking Safe Cut Points for Maximal-Munch Tokenization
SummaryThis paper studies how to certify byte windows in maximal-munch tokenization, where a window is a certified anchor if, in any fully tokenizable input containing it, the token covering its last byte starts at the window's origin. It gives an online method for literal vocabularies and an offline verifier for arbitrary regular token sets. The authors prove that cutting inputs at certified anchors and scanning chunks independently reproduces sequential segmentation, and they supply executable checks and measurements from shipped programs.
WhyOffers decidable anchor certification with theorems on splitting, edits, and differential auditing; executable checks and measured chunked scanning support claims, useful for tokenization.
In briefA simple algorithm that beats the 1/3 approximation barrier for Nash equilibria in two-player games
SummaryComputing Nash equilibria in bimatrix games is PPAD-complete, and this paper studies the strongest approximation guarantee achievable in polynomial time. The authors propose a simple, symmetric, deterministic polynomial-time algorithm with no tunable hyperparameters, achieving an approximation of about 0.30954+δ and breaking the long-standing 1/3 barrier. The method extends the optimization framework of Tsaknakis and Spirakis with building blocks from the authors' work on strategy mixing and automated approximation analysis.
WhyBreaks the long-standing 1/3 barrier for approximate Nash equilibria with a new algorithmic structure and conceptual explanation.
