In briefDiffusion-based secure UAV sensing and communication that hides alert traffic patterns
SummaryThe paper studies a drone that both tracks a mobile target and sends alerts to a ground station when the target enters a protected area. Because these alerts change the communication traffic, an eavesdropper could infer the target's status without decoding messages. The authors propose a keyless scheme in which a conditional diffusion model generates a message-independent safeguard waveform superposed on the data, making alert and non-alert periods statistically indistinguishable. Numerical results with five dB eavesdropper SNR show a learned eavesdropper stays near the 0.5 blind-guess accuracy while the average position PCRB remains between 0.59 and 0.70 square meters.
Homa Nikbakht, Yonina C.Eldar, and H. Vincent Poor
In briefNon-equivalence of mixed-additive and Z_{p^s}-linear generalized Hadamard codes for odd primes
SummaryThis paper studies generalized Hadamard (GH) codes that are Gray map images of Z_pZ_{p^2}-additive codes, comparing them with Z_{p^s}-linear GH codes. The authors show that the kernel blocks of a nonlinear code H_p^{t_1,t_2} yield equivalent linear punctured codes, and they determine the ranks of these codes. For odd primes p, no code H_p^{t_1,t_2} is equivalent to any of the Z_{p^s}-linear GH codes.
In briefTiming and reliability limits of position-assisted UWB multicast audio
SummaryThis paper presents a unified feasibility analysis for one-to-many ultra-wideband audio transport that separates clock error, age-dependent acoustic propagation compensation, correlated block erasures, and reserved communication resources. It derives bounds on single-listener arrival spread and a stationary expression for all-receiver block delivery, validated through 4,590 Monte Carlo runs. Position assistance markedly reduces acoustic arrival spread, yet most compensated events still exceed the 30-microsecond criterion, and the study identifies failure mechanisms such as position age and shared fades.
In briefSubjective Valuations Make Fair Group Allocation Harder: Complexity of Allocating Resources to Agent Groups
SummaryThis paper studies the parameterized complexity of finding EF1 (envy-free up to one item) allocations of indivisible items when items are assigned to groups of agents who share them. It considers both fixed group assignments (EF1-Fix) and variable ones (EF1-Var). The authors show both problems are NP-hard even with a constant number of agents. They also establish fixed-parameter tractability for EF1-Var with respect to the number of subjective items and for EF1-Fix with respect to the number of item types, while linking an analogous result for EF1-Var to an open question in Integer Linear Programming.
Argyrios Deligkas, Eduard Eiben, Robert Ganian et al.
In briefA paper resolving the linear MDS conjecture in even characteristic
SummaryThe paper proves that nontrivial linear [n,k] MDS codes over finite fields of characteristic two satisfy n ≤ q+1. The bound is relaxed to n ≤ q+2 in dimensions k=3 and k=q-1. This settles the linear MDS conjecture for even characteristic.
WhyResolves the long-open linear MDS conjecture in even characteristic with a complete proof, a major theoretical advance.
In briefExact Storage-Repair-Bandwidth Tradeoffs for Exact-Repair Regenerating Codes at (6,5,5) and (7,6,6)
SummaryThis paper gives exact information-theoretic outer bounds that fully characterize the storage-repair-bandwidth tradeoff regions for the parameters (6,5,5) and (7,6,6). The layered code proposed by Tian et al. achieves this tradeoff and is therefore optimal. The proofs are provided as integer-valued certificates along with a Python verifier.
In briefWorst-case Bayesian CVaR regret bounds for episodic MDPs
SummaryThis paper studies the worst-case, over priors, Bayesian conditional value-at-risk of cumulative pseudo-regret in finite-horizon episodic Markov decision processes. A high-probability guarantee for the UCBVI-BF algorithm gives a prior-predictive upper bound. A lower bound is obtained using a three-state MDP and a reverse-KL inequality. For a fixed number of states, the bounds share the same polynomial dependence on horizon, actions, and episodes, differing by a logarithmic factor in the risk level.
Raghav Bongole, Amirreza Zamani, Tobias J. Oechtering et al.
In briefLower bounds on Bayesian CVaR for interactive decisions under local differential privacy
SummaryThis paper proposes an information-theoretic method for lower bounding the prior-predictive conditional value-at-risk of loss in interactive statistical decision making. It uses a generalized Fano inequality and a bounded hinge transform to reduce the problem to a Bernoulli f-divergence inequality, yielding explicit bounds for Gaussian mean estimation and Gaussian stochastic bandits. It then adds local differential privacy at the observation stage and combines LDP contraction inequalities with the adaptive KL chain rule to obtain Bayesian CVaR lower bounds for these settings.
Raghav Bongole, Amirreza Zamani, Tobias J. Oechtering et al.
In briefBalancing Fair Representation Before and After Committee Selection in Structured Elections
SummaryThis paper studies randomized committee elections with additive preferences, aiming for proportionality both in selection probabilities ex ante and in each realized committee ex post. It introduces antichain single-peakedness, a restricted preference domain that generalizes single-peaked additive preferences. The authors compute a fractional committee and convert it into actual committees using systematic rounding. They show ex-ante core stability is incompatible with ex-post justified representation, and establish a tight threshold of αβ=2 for approximate guarantees.
Dimitris Avramidis, Mark de Berg, Ulrike Schmidt-Kraepelin
In briefKPage: A Method for Packing Sparse Attention Pages Across Blocks to Improve Hardware Efficiency
SummaryThis paper addresses sparse attention in long-sequence video Transformers, where 3D token layouts often leave selected blocks partly empty and waste GPU tile capacity. The authors propose KPage, which packs valid 16-row key/value pages across logical block boundaries into larger tiles during loading and routes query blocks between two tile shapes to reduce padding. The reported results include 1.35x to 2.06x forward-kernel speedup over NVIDIA Block-Sparse Attention, and speedups on eight layouts from other open-source sparse attention systems.
In briefDeterministic causal quantization using shared representatives with bounded error
SummaryThe paper studies deterministic causal quantization that keeps pathwise error at most ε, where shared representatives are updated using only transmitted addresses. It compares address entropy rates with the process ε-entropy and bounds the gap for stationary sources with finite scalar ε-entropy. The authors also present results for uniform lattice grids, autoregressive Laplace and Gaussian sources, and stationary processes on finitely many centers.
In briefLength bounds for irreducible cyclic MDS codes
SummaryThe paper proves the MDS length bound for nontrivial irreducible cyclic codes of every composite length, using trace realization to convert the MDS property into independence of subgroup elements. A cyclotomic argument reduces each fixed dimension to finitely many parameters, and the standardness conclusion is stated only as a conditional consequence. The general prime-length problem is not settled.
WhyReduces MDS property to independence in multiplicative subgroups, proving length bounds for all composite lengths; prime-length case remains open.
In briefError-correction decoding for 2D convolutional codes using a generalized Viterbi algorithm
SummaryThis paper addresses the lack of practical decoding methods for two-dimensional convolutional codes, which are suited to encoding 2D data. The authors introduce a 2D Viterbi algorithm that extends the classical one-dimensional Viterbi approach to the two-dimensional setting. The algorithm operates on linear systems representations of 2D convolutional codes, and its complexity depends on the dimension of the corresponding 2D linear system. The paper also provides several methods for obtaining small-dimension 2D realizations.
In briefDividing Planar Point Sets Fairly with Few Straight Lines
SummaryThe classical necklace splitting problem asks how to divide beads of different colors among several people with as few cuts as possible, so each person gets equal amounts of each color. This paper studies a planar version where cuts are straight lines separating red and blue points. It presents static and dynamic algorithms that narrow a known gap in the number of lines needed. A general algorithm uses at most k + log₂(k+1) − 2 lines and runs in Θ(m log k) time.
In briefinfomeasure-rs: A High-Performance Rust Library for Information Measures
SummaryThis paper presents infomeasure-rs, an open-source Rust library that combines many information-theoretic measures, such as entropy, mutual information, transfer entropy, and divergences, with several estimator types in one portable, type-safe package. On a single CPU core, it is the fastest of the compared cross-language packages, with especially large gains for mutual information, conditional and transfer measures, and several estimators. Optional GPU and multi-threaded paths further speed up some continuous estimators.
In briefBenchmarking CPU, GPU, and ML Power-Flow Solvers from 14 to 70,000 Buses
SummaryThis paper compares seven open-source CPU power-flow solvers on identical inputs and finds throughput varies by roughly 2.5 to 3 orders of magnitude, so the chosen baseline can change the conclusion. The authors build a benchmark of nine physics-based solvers, including a new batched GPU solver called gpusim2grid, and three published ML surrogates across 15 grids from 14 to 70,000 buses. They report that a well-parallelized CPU solver or gpusim2grid matches or exceeds the surrogates' throughput on power-flow while being orders of magnitude more accurate.
WhyUnified benchmark of nine solvers and three ML surrogates across grids up to 70,000 buses shows baseline choice can flip conclusions.
Benjamin Donnot, Sami Tazi, Eva Boguslawski et al.
In briefAlmost entropic matroids and (k,L)-systems explained
SummaryThe paper studies (k,L)-systems, families of k-element sets in which any two distinct members intersect in a size belonging to the set L. It defines m(n,k,L) as the maximum size of such a system on an n-element ground set. The authors connect the asymptotic growth of m(n,k,L) to almost entropic matroids. This yields a general computable upper bound that unifies and strengthens all previously known bounds.
WhyLinks (k,L)-system asymptotics to almost entropic matroids, yielding a unified, computable upper bound that strengthens prior results.
In briefExtending Mechanisms: Generalizing Incentive-Compatible Mechanisms from Restricted Valuation Domains
SummaryThe paper studies how an incentive-compatible mechanism for a single agent, designed on a restricted domain of valuations, can be extended to the entire valuation space while preserving incentive compatibility. It identifies two extreme extensions. The "keep-the-menu" extension minimizes buyer payoffs and maximizes canonical prices pointwise, while the "keep-the-pricing" extension does the opposite.
In briefTesting runtime prediction and safe walltime limits for supercomputer jobs
SummaryThis paper introduces SAFE, a framework that works with any runtime predictor and turns predicted runtimes into conservative, bounded walltime recommendations for HPC jobs. Its main contribution is this decision layer rather than a new prediction model. Tests on production job histories from Fugaku and NREL Eagle show that the approach reduces walltime gaps, while keeping job termination rates below 1% on both systems.
Hector Romero Ugalde, Youssef Faqir-Rhazoui, Marlon Funk et al.
In briefEfficient perturbation-based SC decoding for long polar codes
SummaryThis paper analyzes how adding perturbations to soft information improves successive cancellation (SC) decoding of polar codes. It shows the gain comes mainly from correcting the first erroneous bit estimate, without weakening SC's error-correction ability. Based on this, the authors design optimal perturbations and propose a hybrid perturbation-based SC (HPSC) decoder combining random and biased perturbations. Simulations indicate that HPSC with optimal perturbation performs similarly to CRC-aided SC list (CA-SCL) decoding while avoiding path sorting and expansion.
In briefInformation-estimation relations for discrete channels via Poisson observation counts
SummaryThe paper studies arbitrary discrete memoryless channels with finite alphabets by varying the expected number of observations while keeping the channel fixed. It shows that the derivative of mutual information with respect to this observation budget equals the minimum expected relative-entropy loss for estimating the input's output distribution, achieved by the posterior-averaged predictor. The results extend to finite-state processes sampled at Poisson times and to the coupon-collector channel, where exact capacity loss is derived.
In briefConstrained federated reinforcement learning for secure task offloading in vehicular edge computing
SummaryThis paper proposes CRAFT, a framework for offloading dependent vehicular computing tasks to roadside edge servers. It models DAG-structured offloading and resource allocation as a constrained Markov decision process, using Lagrangian relaxation to enforce task-completion requirements and two-tier federated aggregation to handle heterogeneous data. Federated parameter exchange is protected by post-quantum homomorphic encryption and lattice-based signatures. On a testbed over a Melbourne CBD network, the authors report up to 42% lower weighted cost, about 96% completion rate, and 6.4% security overhead.
In briefAoI-aware polar code rate adaptation without channel estimation
SummaryThis paper studies how to adaptively choose how many of the most reliable coordinates to use in polar-coded transmission, balancing age of information against throughput. The controller updates the prefix length using only the last decoding outcome, with no channel estimation. The authors derive a closed-form stationary distribution and exact steady-state AoI and goodput, and simulations on a 64-coordinate polar code over a binary symmetric channel are used to check these predictions. The paper also quantifies the goodput cost of adapting from one feedback bit compared with a genie-aided bound.
In briefWhether the simple strategy-iteration method with Dantzig's rule runs in strongly polynomial time on turn-based deterministic games
SummaryThis paper studies turn-based deterministic zero-sum games with discounted and long-run average costs. For discounted costs, it gives exponential and superpolynomial lower bounds on iterations for the simple strategy-iteration method with Dantzig's rule, under unrestricted costs and unit weights respectively. For long-run average costs, it shows an exponential iteration lower bound for a Dantzig-based counterpart with unrestricted costs, and proves it has a strongly polynomial implementation with unit weights.
WhyProves exponential and superpolynomial lower bounds for Dantzig-rule strategy iteration, refuting strong polynomiality even with costs in {-1,0,1}.
In briefSionna Bench: A Benchmark Framework for Wireless Communications Research
SummaryThe paper notes that wireless communications research has lacked benchmarks, so methods were developed for slightly different setups and are hard to compare. The authors present Sionna Bench, starting with MIMO-OFDM channel estimation as a pilot. Using AI-assisted coding, they implemented twenty channel estimation methods from ten papers within a common comparison framework.
Gian Marti, Jakob Hoydis, Fay\c{c}al A\"it Aoudia et al.
In briefBlock basis reduction for linear codes under non-Hamming distance measures
SummaryThis paper extends code basis reduction to non-Hamming metrics by proposing a block basis reduction (BKZ) algorithm that applies to all coordinate-additive metrics. It generalizes the Ghentiyala and Stephens-Davidowitz algorithm. For block size 2 (LLL), the authors give a polynomial-time algorithm with complexity O(n^4 q^3) operations over F_q. They also implement the method for the Lee metric and report experiments on finding low Lee-weight codewords.
In briefTight Hardness Threshold for Approximating Generalized Circuits
SummaryGeneralized circuits are a key tool for proving PPAD-hardness and membership for approximate equilibrium problems. The paper establishes a tight threshold at ε=1/6: ε-GCircuit is PPAD-complete for every constant below 1/6, even with four gate types, and polynomial-time solvable for constants at or above 1/6. It closes the gap between earlier bounds of 1/10 and 1/2. The authors also study a relaxation allowing a fraction of gates to be violated, and apply the results to problems in algorithmic game theory.
WhySettles the open threshold for generalized circuit approximation at ε=1/6 with matching PPAD-hardness and polynomial-time proofs, plus a decomposition theorem with broad game-theory applications.
In briefFaaSpot: Fault-tolerant serverless workflow execution on cheaper spot instances
SummaryThis paper presents FaaSpot, a system that runs serverless DAG workflows entirely on spot instances, which are cheaper but can be preempted. It analyzes each workflow's topology at submission time to classify tasks by how severely their preemption would cascade, then applies selective checkpointing, survival-aware scaling, and lineage-based retry. In trace-driven evaluation on eight DAG workflows, FaaSpot completed 89.03% of runs at $0.018 per successful run and reduced mean latency by 2.3x.
Bandhan Patel, Sushruth Harsha, Kwangsung Oh et al.
In briefUpper bounds on service rates of linear codes and links to combinatorial designs
SummaryThe paper studies aggregate service rates, the concurrent demands a distributed storage system can serve without overloading servers, for systems built on linear codes over arbitrary finite fields. It derives an upper bound for any subset of data objects using the dual distance and smallest recovery set sizes, which for systematic generator matrices depends only on code length, dual distance, and subset size. The authors give conditions under which the bound is attained and apply the results to MDS, Reed-Muller, constacyclic, and extended Golay codes to obtain exact maximum aggregate service rates.
In briefA study of relative Feng-Rao distances, focusing on C_ab curves
SummaryThis paper studies two kinds of mth relative Feng-Rao distances, called primary and dual, which lower-bound relative generalized Hamming weights of nested one-point codes. The authors develop a unified method yielding generic formulas for general numerical semigroups and closed formulas for many semigroups generated by two elements, complementing earlier work by Delgado et al. For non-singular C_ab curves, they also propose a method to estimate these weights where generic methods may fail.
Eduardo Camps-Moreno, Olav Geil, Hiram H. L\'opez et al.
In briefCQTwin: A digital-twin toolkit for analyzing supercomputer job scheduling
SummaryThis paper presents CQTwin, an open-source toolkit that creates digital twins of HPC job schedulers to analyze operational data. It offers offline trace replay of archived workloads and online event-driven simulation of queued jobs, built on the CQSim simulator. An interactive dashboard lets users characterize workloads, compare scheduling policies, and explore what-if scenarios. The toolkit is validated using production traces from the Argonne Leadership Computing Facility from 2016 to 2023.
In briefBayesian analysis of transaction fee mechanisms when multiple proposers act concurrently
SummaryThis paper studies trade-offs in Bayesian equilibria of transaction fee mechanism design when n proposers propose blocks concurrently. The authors develop analytical tools that examine sequences of user-proposer equilibria as the number of users grows while holding relative congestion fixed. They compare the pay-once and pay-per-inclusion (PPI) mechanisms, finding that PPI is efficient in the limit while pay-once remains inefficient, and that exclusive proposer payments yield inefficient equilibria.
WhyIntroduces a new limit-sequence toolkit proving PPI efficiency and asymptotic inclusion behavior, a notable theoretical contribution.
In briefBinomial distribution maximises entropy among negatively related indicator sums
SummaryThe paper shows that the binomial distribution on {0,1,…,n} with mean np (0<p<1) maximises entropy among sums of negatively related indicator random variables with the same mean and support. This generalises Yu (2008), which covered only the smaller class of ultra-log-concave random variables of order n. The proof uses a Markov chain construction from Yu (2008) together with a stochastic ordering argument.
