In briefFormation of environmental agreements and payoff sharing among countries.
SummaryThe paper addresses how countries must bear costly emission reductions for climate change, yet each prefers others to act while benefiting from a cleaner environment. It notes that game theory suits designing international environmental agreements, and that prior work treats coalition formation noncooperatively while cooperative games are used to share climate costs. Both approaches are described as having advantages and drawbacks.
Michel Grabisch, Elena Parilina, Agnieszka Rusinowska et al.
In briefMovable-antenna anti-jamming: provably locally optimal design and a unified antenna-position optimization framework
SummaryThis paper studies how movable antennas (MAs) can improve anti-jamming in wireless receivers by jointly optimizing antenna positions and the receive beamformer to maximize output SINR. For a single jammer and single line-of-sight path, it proposes a damped projected-gradient descent (DPGD) algorithm whose complexity scales linearly with the number of antennas and proves it converges almost surely to a local minimizer. For general cases with multipath or multiple jammers, it proposes a soft-active projected-gradient (SAPG) algorithm, also linear in complexity, applicable to MVDR and ZF reception. Numerical results show gains in anti-jamming performance and computational efficiency.
In briefA short note on how numerical error bounds change with hardware resources
SummaryThis short technical note explains how to bound numerical errors relative to computational resources. It combines known complexity estimates for parallelizable algorithms with parallelization laws. The authors suggest the approach generalizes beyond their examples and can fairly compare algorithms, hardware, and software by giving each configuration the same resource or energy budget.
In briefA unified framework for product assortment optimization and matching under multinomial logit customer choice
SummaryThe paper develops a unified framework for assortment optimization and matching when customers choose according to the multinomial logit (MNL) model. It centers on a single-customer problem where the firm commits to eligible products before their availability is revealed. A simple threshold policy achieves a tight 1/2 guarantee against a benchmark and accommodates cardinality constraints. Applied to several problems, the framework yields 1/2 guarantees and a 1/4 guarantee for fully static two-sided assortment optimization.
WhyProposes a unified framework with tight 1/2 guarantees improving several applications' approximation factors, mainly theoretical in impact.
In briefConverse of symplectic interlacing via low-rank singular value perturbations
SummaryThe paper proves the converse of the symplectic interlacing theorem. Given a global symplectic spectrum and a local symplectic spectrum satisfying the 2k-step interlacing inequalities, it constructs a real symmetric positive definite matrix with these prescribed spectra, where the local spectrum belongs to a principal submatrix. The main tool reduces s-principal compression to a rank-k singular value perturbation. In the Gaussian quantum marginal problem, the symplectic interlacing inequalities are shown to be the complete spectral compatibility conditions for the global-versus-local-mode marginal problem.
WhyProves the converse of the symplectic interlacing theorem with a complete characterization via low-rank singular value perturbations, a notable theoretical advance.
Priyanka Grover, Hemant K. Mishra, Sarishti Singh et al.
In briefMinimality of binary codes built from Hamming weight layers and graph cut spaces
SummaryThis paper studies the minimality of binary linear codes, which matter for secret-sharing schemes. It examines three known constructions: codes from a single complete Hamming-weight layer, codes from unions of such layers, and graph cut-space codes. For single-layer codes it gives a complete classification, showing they are minimal exactly when t is even. For unions of layers it gives an if-and-only-if minimality criterion, and for graph cut-space codes it translates minimality into a graph-theoretic condition.
In briefFurther Study of p-Ary Self-Orthogonal Codes from Square Functions
SummaryThis paper studies a family of trace codes over finite fields of odd characteristic, defined by two quadratic trace conditions. By analyzing the associated quadratic forms and evaluating character sums, it determines the codes' parameters and complete weight distributions. It also builds punctured codes using the multiplicative orbit structure of the defining sets and establishes self-orthogonality criteria, covering divisibility and projectivity.
In briefWho Pays for Hidden Reasoning? Billing Model Cloud Tokens
SummaryThis paper studies how charging for hidden internal reasoning and candidate answers in token-priced language model services affects provider decisions and social welfare. It builds a three-party model of users, a model firm, and an execution cloud, examining who controls the computation budget. A LiveCodeBench study compares answer quality, latency, token use, and resource consumption across five model deployments. It finds metered pricing yields higher welfare when the extra inference creates net social value, while fixed pricing is preferable when its costs outweigh benefits.
In briefBROCADE: A broker-based contract design for multi-agency welfare maximization.
SummaryThis paper studies welfare maximization in a contract model with m principals who privately value a public project's success and n agents whose effort is unobservable. It introduces three tools: a broker that separates principal payments from agent payments, an extension of Myerson's Lemma for dominant-strategy incentive compatibility (DSIC), and a black-box reduction to budget-constrained single-principal contract design. The authors give approximation results for deterministic and randomized contracts under different benchmarks, including a universally DSIC and budget-balanced randomized contract with an O(log m) approximation ratio independent of n.
In briefSharp limits on graph codes avoiding complete bipartite differences
SummaryThe paper studies how large a fraction of all graphs on n vertices can be chosen so that no two chosen graphs differ by exactly a complete bipartite graph K_{s,t}. For every pair s,t the authors determine this maximum density. When both parts are odd it is 1/2; otherwise, with 2h the largest even part size, the density is Θ(n^{-h}). The upper bound uses the Frankl–Füredi forbidden-intersection theorem, and the lower bound uses BCH-based binary linear codes.
WhyDetermines the exact order Θ(n^{-h}) for all fixed s,t, improving Alon's O(1/n) bound via Frankl–Füredi upper bound and BCH-code lower bound.
In briefA revisit of entropy and related measures for symbolic sequences and how to estimate them
SummaryThis paper revisits entropy, entropy density, effective measure complexity, and informational distance for analysing sequential data, grounding them in information theory and ergodic theory. A central result, Ziv's theorem, shows that the normalised number of distinct words is asymptotically equivalent to entropy density, supporting estimation via Lempel-Ziv factorisation. The authors introduce estimation procedures for finite sequences and an open-source parallel software library. Three case studies illustrate use: functional brain connectivity from fMRI, coupled nonlinear oscillators, and ECG-based atrial fibrillation classification.
Efren Aragon-Perez, Ania Mesa-Rodriguez, Karel Garcia-Medina et al.
In briefNew additive non-RS MDS and ACD codes from twisted Reed-Solomon codes
SummaryThis paper studies single-twist additive generalized twisted Reed-Solomon (GTRS) codes over finite fields. It gives necessary and sufficient conditions for these codes to be MDS for arbitrary hook and twist parameters. Using Schur squares, the authors show such additive MDS codes are not monomially equivalent to additive Reed-Solomon codes. They also identify a class of additive GTRS codes whose duals are again additive GTRS codes, which yield ACD codes for sufficiently large prime powers with characteristic avoiding a finite set.
In briefA priority-aware scheduler that reduces cold starts and balances load for serverless functions
SummaryThis paper presents SPES-Ext, a rule-based, priority-aware scheduler for memory-constrained distributed worker nodes running serverless functions. It categorizes functions by their invocation patterns and combines predictive function provisioning, eviction optimization, and dynamic load balancing. According to the authors' experiments, it cuts 75th-percentile cold starts by 80%, reduces memory waste by 83.92%, and improves worker load variance by 52.49% over state-of-the-art solutions.
In briefWhether Follow the Regularized Leader fails to converge in constrained optimization
SummaryThe paper studies whether Follow the Regularized Leader (FTRL), a basic online learning algorithm, converges to first-order stationary points in constrained nonconvex optimization. The authors construct an infinitely smooth constrained problem where continuous-time FTRL's KKT gap stays bounded away from zero indefinitely. They also show that FTRL with Euclidean regularization converges when the objective function is analytic.
WhyConstructs an infinitely smooth problem where continuous-time FTRL keeps a large KKT gap, a surprising theoretical result with proofs.
Ioannis Anagnostides, Ioannis Panageas, Nikolas Patris et al.
In briefCounting subcodes of linear codes with prescribed hull dimension, plus containment and limiting distributions
SummaryThe paper counts subcodes of a fixed parent linear code having prescribed dimension, hull dimension, and type over finite fields. It splits the parent code into an LCD complement and its hull, and uses projection to reduce the problem to the LCD complement, yielding explicit formulas. The authors also treat the dual containment problem and derive asymptotic distributions of hull dimensions as the relevant dimensions grow.
In briefFrontGI: Architecture-Aware Parallel GNN Inference on Chiplet Multicore Systems
SummaryGraph Neural Network inference on large graphs often scales poorly on modern multicore systems with over a hundred cores when using standard libraries. This paper presents FrontGI, a set of techniques that choose parallelization strategies based on graph structure and multicore cache hierarchies. Evaluated on a 128-core AMD EPYC system, FrontGI achieves up to 3.35x, 3.20x, 2.89x and 3.10x speedups over DGL for GCN, GraphSAGE, GIN and SGC, with larger gains over PyG.
In briefWhen and how to use NVIDIA Hopper's TMA for GPU memory loads, based on real kernel workloads.
SummaryThis paper studies when GPU kernels should use the Tensor Memory Accelerator (TMA) on NVIDIA Hopper rather than alternatives such as cp.async or direct global loads. The authors catalog the load choices made in FlashAttention-3 and use microbenchmarks on H100 to identify transfer-size crossovers and multicast scaling effects, yielding rules of thumb. They also compare TMA multicast with distributed shared memory and check, via a CUTLASS GEMM study on B200, that the dispatch rules hold qualitatively on server Blackwell (SM100).
Christin David Bose, Cesar Avalos, Junrui Pan et al.
In briefFair Division of Uncertain Items (Lotteries)
SummaryThis paper studies fair division of uncertain items, where each item's utility is a random variable revealed only after allocation. It distinguishes ex-ante fairness, assessed before utilities are realized, from ex-post fairness, assessed afterward, and examines how compatible the two are. The authors identify conditions under which ex-ante envy-free allocations maximize ex-post envy-freeness probability and analyze approximation guarantees and computational complexity.
In briefOrder-selection prophet inequality nearly matches the i.i.d. bound
SummaryThe paper studies the order-selection prophet inequality, where a gambler who knows the distributions of independent nonnegative values chooses the inspection order and may irrevocably accept at most one value. It presents an algorithm with competitive ratio 0.74500001, less than 4.5×10⁻⁴ below the optimal i.i.d. ratio of about 0.74544. The guarantee combines analytic estimates with a rigorous computer-assisted certificate.
In briefConstructing Secure Convertible Codes for Secrecy Capacity
SummaryThis paper constructs two families of access-optimal secure convertible codes. They achieve secrecy capacity under a passive eavesdropper model that can access any ℓ symbols from the initial and final codewords. The first family, based on Maximum Rank Distance (MRD) codes, covers all parameters in the merge and split regimes but needs a large field size. The second family, based on Generalized Reed-Solomon codes, uses smaller fields but only applies to certain parameters in those regimes.
In briefHiLO: A Framework for Product-Level Graceful Degradation Under Heavy Load
SummaryThe paper presents HiLO (High Load Optimizer), an automated system that maps observed load to a lattice of pre-validated product states ordered by degradation severity. Dependencies are resolved at design time, allowing deterministic, continuous control without runtime dependency discovery. According to the paper, HiLO is deployed in Microsoft Bing, where it keeps the OneDCG relevance drop under 0.5 points even under sustained load exceeding 2x baseline peak.
Knut Magne Risvik, Zhuowei Li, Nirupam Purushothama et al.
In briefNonexistence and Near-Constructions of Complete Coverings of Lee Spheres
SummaryThe paper studies complete coverings, finite configurations of pairwise disjoint radius-r Lee spheres around codewords of perfect Lee codes in Z^n. It proves that no complete covering exists when n≥8 and r²+⌊25r/34⌋−3≥2n, and gives an improved large-radius nonexistence bound for perfect Lee codes. It also constructs almost complete coverings using near-optimal ternary constant-weight codes.
WhyIntroduces complete coverings and substantially strengthens nonexistence bounds for perfect Lee codes with rigorous proofs; practical impact is limited.
In briefLinear Hashing Is Poissonesque: Bin Loads Under Random F2-Linear Maps
SummaryThe paper studies how m balls distributed into n bins behave when placed by a random F2-linear map rather than fully at random. It shows the load distribution is still governed by a Poisson distribution of mean m/n, though with a weaker guarantee. The authors conclude that linear maps closely match fully random maps in expected loads and concentration, with bounds on maximum load and on loads of bin subsets.
WhyProves linear hashing loads follow Poisson bounds, with max-load gap to random maps at most 3/2+o(1), a rigorous theoretical advance.
SummaryThis paper studies the rank, meaning the dimension of the binary linear span, of Z_{2^s}-linear Hadamard codes, which are Gray images of Z_{2^s}-additive codes. Previously the rank was known in closed form only for s≤3, and for s≥4 only through computer results for small lengths. The authors give a method that computes the rank for every s and show it is a binomial polynomial in type parameters with length-independent coefficients. They compute all coefficients for s≤5, obtaining explicit rank formulas over Z_16 and Z_32 as well as recovering the known ones over Z_4 and Z_8.
SummaryThis paper proposes a semantic successive cancellation list (SCL) decoder for polar codes that carry natural-language text. A fine-tuned byte-level language model supplies next-byte priors, which are converted into bit costs and added to the SCL path metric. Simulations on length-512 polar codes show large block error rate improvements over CRC-aided SCL. At rate 0.75 and list size 16, the gain exceeds 1 dB compared with CA-SCL at the same list size.
In briefWhich parts of a generative rate-distortion-perception system can be reused for a new task
SummaryThe paper studies generative lossy compression systems and asks what remains reusable when one is transferred to a new conditional-generation task. It compares a full redesign with three transfer strategies of increasing restriction. For finite alphabets it characterizes exact rate regions, controllability conditions, and compression-rate penalties, and it gives closed-form results for a scalar Gaussian case.
In briefLive Adaptation of Dataflow Stream Processing for Hybrid Edge-Cloud Analytics
SummaryThe paper argues that neither all-cloud processing nor static edge offloading can reliably meet target rates under dynamic WAN conditions and limited edge resources. It presents Sluice, an adaptive stream processing system for hybrid clouds that is compatible with dataflow engines such as Apache Flink, featuring an automatic query rewriter and a live routing mechanism that moves computation between edge and cloud without pausing execution. Evaluated on a Raspberry Pi testbed and a 100-VM simulated testbed across five real-world applications, Sluice is reported to shift computation seamlessly and achieve consistently higher throughput than baselines.
In briefCompetitive exploration in structured environments: a minimal model of two foragers competing over a distant reward
SummaryThis paper studies how competition changes the balance between exploring and exploiting for foragers who know a small nearby reward but not a larger distant one. The authors use a minimal model of two epsilon-greedy temporal-difference learners on a one-dimensional lattice. They find that the Nash equilibria include both exploring, both staying, or one exploring while the other stays, and that local best-response dynamics often converge on outcomes that are socially suboptimal. A minimal signal in which one forager reports finding the distant reward can remove many of these dilemmas.
In briefA Method for Sampling and Recovering Tensor-Valued Dynamical Systems
SummaryThis paper studies how to sample and recover tensor-valued systems that evolve under the t-product, using partial spatial observations over time. It gives deterministic conditions under which the initial tensor state can be identified, and a two-stage recovery method with high-probability guarantees and stability estimates. Experiments on synthetic, sea surface temperature and turbulence velocity data show that the learned t-product operator generally outperforms several baseline methods.
In briefVerifiable, failure-resilient distributed approximate nearest-neighbor search with per-query trust checks
SummaryThis paper presents TrustANN, a middleware layer that adds query-level trust to distributed approximate nearest-neighbor vector retrieval without changing the underlying ANN algorithm. It marks a query as Trusted only when every relevant shard has enough independent, authenticated, consistent replica evidence, and otherwise returns Unverified. On an AWS deployment using distributed FAISS HNSW, the authors report that attack success dropped from 1.000 to 0.000 and clean p95 latency rose by 9.36%. The system also maintained trusted retrieval with up to 33% replica failures.
In briefCo-locating Cross-Modal Data Across Heterogeneous Storage
SummaryThis paper presents Coloc, a storage-management layer that treats cross-modal access locality as a primary signal for data placement. It uses a Cross-Modal Access Graph combining semantic affinity, co-access, and temporal behavior to place frequently co-accessed vectors under capacity and topology constraints. Deployed on a multi-tier AWS testbed spanning GPU, NUMA-DRAM, NVMe, and S3, Coloc substantially reduces data movement and query latency while leaving ANN retrieval unchanged.
In briefSupervision Without Blocking: Redesigning OTP-Style Fault Tolerance for Cooperative Runtimes
SummaryActor frameworks on mainstream runtimes claim Erlang/OTP-style supervision, but OTP's guarantees assume a synchronous, single-threaded supervisor. The authors checked OTP's documentation and source, and compared three Rust actor frameworks, identifying 24 defect classes, eight of which share one root cause. They present a supervision runtime that closes all 24 classes, validated by 241 tests, and keeps OTP's verified contracts. It also adds two properties: the supervisor stays responsive during group restarts, and message handlers are never cancelled mid-execution.
In briefDecoding algorithms for insertion and deletion errors in MDS codes
SummaryThis paper studies decoding maximum distance separable (MDS) linear block codes when insertion and deletion errors occur. The authors present a bounded distance decoding algorithm for any MDS code of rate at most one half, using a sliding window with lower time complexity than brute-force decoding. For optimal Reed-Solomon codes, it corrects insertion-deletion errors up to the half-Singleton bound. The paper also gives two faster algorithms for the special cases of insertion-only and deletion-only errors.
In briefPrefix Keys and Values Matter Late: RIPE Caching for Looped Models
SummaryThis paper studies the key-value cache of looped language models, which run the same layer block several times per token. It finds that later loops are far more sensitive to prefix keys and values than earlier ones. The authors propose RIPE, which stores only the deepest cache plane except for a window of recent positions kept at full depth. Experiments on Huginn and Raven-Llama report self-speculative decoding speed-ups and memory savings.
In briefEnvy-free allocations under random non-additive valuations
SummaryThis paper studies fair division in a semi-random model where each agent has an arbitrary base valuation over bundles and uniformly permutes item names at random. For any fixed number of agents n and any number of items m divisible by n, the authors show that an envy-free allocation exists with probability 1-o(1) as m grows. This resolves an open question of Benade et al. for n greater than 2. The proof uses the second moment method together with the (small-set) expansion of generalized Johnson graphs.
WhyResolves an open question for all agent counts by combining second moment method with generalized Johnson graph expansion, yielding a rigorous proof.
In briefMANTA: Making DAG-Based Asynchronous BFT More Flexible
SummaryThis paper studies DAG-based asynchronous Byzantine fault-tolerant consensus protocols. Existing designs suffer long rounds on heterogeneous wide-area networks because replicas wait for synchronized rounds, which slows them down when stragglers exist. The authors propose a flexible DAG growth framework that relaxes round synchronization and accumulates the connectivity needed for commitment across rounds. The resulting protocol, MANTA, reduces latency by roughly 30% to 70% in experiments on heterogeneous WAN deployments.
In briefA 28 GHz path-level geometric-stochastic channel model for ISAC target scattering, derived from measurements of robots, humans and cars
SummaryThis paper builds a scattering channel model for integrated sensing and communication (ISAC) targets, including robots, humans and automobiles, using a unified 28 GHz measurement campaign. Nine targets were measured across many transmitter-receiver configurations, yielding nearly 117,000 acquisitions. Multipaths were extracted with a phased-array sounder and super-resolution processing, and the cluster center is set by bistatic specular geometry while its shape is modeled stochastically. Measured and synthesized channels show close agreement in pathloss trends and fluctuation behavior.
In briefLogarithmic asymptotics for counting polynomial threshold functions
SummaryThe paper counts Boolean functions on the n-dimensional cube that are the sign of a real polynomial of bounded degree. It shows that the logarithm of this count is asymptotically equal to that of the classical hyperplane-arrangement upper bound, with relative error O((log n)^2/n), uniformly across all degrees. Sharper error estimates are also obtained in certain degree ranges, using tools such as probabilistic estimates for Reed–Muller codes and an entropy count.
WhyMatches logarithmic counts of polynomial threshold functions to hyperplane bounds with relative error O((log n)^2/n), a clear theoretical improvement via full proofs.
In briefMinimum field extension degrees needed for exact Reed-Solomon code repair at the cut-set bound
SummaryThe paper studies how large a field extension a Reed-Solomon code needs so that every lost node can be exactly repaired from any chosen set of d helpers, at the cut-set bound. For 3≤k<d<n≤2d-2, the authors prove the minimum extension degree over any prime power field is exactly sQ_s(n), where s=d-k+1. They also show that any nonlinear repair scheme reaching the cut-set bound induces an F_p-linear scheme, so the prime-field lower bounds hold without assuming linearity.
WhyGives exact minimum extension degrees for Reed-Solomon repair at cut-set bound and proves nonlinear repair reduces to linear.
In briefPricing Information in Search with Optional Inspection: How platforms can charge for information when buyers may pick options without buying reports
SummaryThe paper studies how platforms can price information that aids decisions as AI agents reshape Web search and advertising. Buyers may select any alternative without purchasing its report. It shows that a common fee maximizes static revenue for identical priors, while exact optimization is NP-hard for heterogeneous binary priors. A polynomial-time static pricing algorithm achieves a 3-approximation to information surplus.
WhyProves uniform fee optimality for identical priors, resolving a conjecture, and gives NP-hardness plus a tight logarithmic approximation bound.
