In briefQuery Complexity of Envy-Free Fair Division for Cake, Chores, and Mixed Goods
SummaryThis paper studies how many queries are needed to achieve envy-free division of cake, chores, and resources with both positive and negative values. It gives a deterministic exact protocol for arbitrary nonnegative nonatomic chore costs, and Las Vegas protocols for cake and chores with polynomial expected query counts. It also shows that exact protocols generally need Ω(n²) expected queries, and that no uniform finite expected-query bound exists for mixed cake and chores.
In briefJoint communication and control design for center-fed pinching-antenna systems
SummaryThis paper proposes an analytical and optimization framework for jointly designing communication and control in a center-fed pinching-antenna system (C-PASS). It characterizes the coupling between data throughput and closed-loop control cost and establishes an achievable communication-control performance region. A throughput maximization problem is formulated to jointly optimize digital and pinching beamforming under control-cost constraints, and solved with an alternating optimization algorithm. Numerical results show 110.2% higher frame-average throughput than conventional PASS, and up to 66.7% gain of concurrent data-control transmission over separated transmission.
In briefSet reconciliation with linear sketches needing only O(d) extra bits
SummaryThis paper studies set reconciliation, where two parties recover the symmetric difference of their sets from short F_2-linear sketches. The authors construct a randomized sketch of length L(n,d)+O(d) plus lower-order terms, with polylogarithmic update time and near-linear decoding time. The construction encodes permuted rows with randomly scaled Pinsketch syndromes, sketches the bit array column by column, and decodes using a recursive sketch called HintSketch.
WhyThe authors achieve L(n,d)+O(d) bits for linear set reconciliation via randomly scaled Pinsketch syndromes and column-wise hinted decoding, with stated polylog update and near-linear decoding time.
In briefRelief: a per-step load-balancing engine for expert-parallel MoE serving
SummaryIn expert-parallel Mixture-of-Experts serving, each layer waits for the most loaded GPU in its group, and routing shifts at every decode step. The paper introduces Relief, a bidirectional pairwise relaxation that avoids exact linear-program solving and reaches within a fraction of a point of the per-step optimum in a few integer-halving passes. The authors measure three deployments: CUDA kernels, an FPGA over PCIe, and a fixed-function block beside the accelerator. They report that the kernel alone matches or exceeds every deployable scheme in effective utilization across group sizes.
In briefA simpler Phase King protocol that uses predictions about faulty processes
SummaryThis paper studies synchronous binary Byzantine agreement in a learning-augmented setting, where each process receives predictions about which processes are honest or Byzantine. The authors present a deterministic, non-authenticated Phase King protocol that combines early stopping with a prediction-aware king selection procedure. For fewer than n/3 Byzantine processes, it tolerates arbitrarily many prediction errors and terminates within O(P) rounds, using O(n²P) messages.
Antonia Giannoulea, George Papadoulis, Nikos Leonardos et al.
In briefGenerating AWS Serverless Infrastructure from WebAssembly Component Interfaces
SummaryThis paper proposes an infrastructure-from-code approach that provisions cloud resources from WebAssembly Interface Type (WIT) definitions. The authors build two interpreters: one derives S3 bucket infrastructure-as-code definitions and least-privilege IAM permissions, while the other runs the component as an AWS Lambda function and grants S3 access as capabilities. The work formulates consistency properties ensuring bucket existence and correct permissions, and shows that the technique prevents selected deployment failures that a baseline setup does not.
In briefImproving bounds on hash and trifferent codes
SummaryThe paper improves probabilistic lower bounds on the rates of hash codes and list-recoverable codes. For perfect q-hashing it proves R_q > B_q for all q ≥ 3, and for non-perfect hashing it proves R_{q,3} > B_{q,3} + 1/250000 for all q ≥ 4. The case q = 9 combined with the Tetra code strictly improves the Körner-Marton lower bound for trifferent codes. The method is also extended to (p,2,3)-list recovery.
WhyThe authors improve probabilistic bounds for hash and list-recoverable codes, proving strict gains over known rates for several parameter regimes, including trifferent codes.
In briefWelfare in envy-free contracts: when agents have few interests
SummaryThe paper studies welfare maximization under envy-free contracts, where a principal assigns tasks and offers success-based payments to heterogeneous agents with hidden effort. For unit-demand agents, the problem is polynomial-time solvable when each agent is interested in at most two tasks, but NP-hard with three. When agents may receive multiple tasks, no polynomial-time constant-factor approximation exists unless P=NP.
In briefDesign properties of doubly-extended Reed-Solomon codes over q-ary alphabets
SummaryThis paper studies doubly extended Reed-Solomon codes with parameters [q+1,k,q-k+2]_q. It shows that for 2≤k≤q-1, the minimum-weight codewords do not form a q-ary 3-design, and extends this to every nonempty weight class with 3≤w<q+1. It also proves that each nonempty positive-weight class is a q-ary 2-design exactly when gcd(k-1,q-1)=1.
In briefQoS-aware energy-efficient near-field hybrid MIMO for multiuser sub-THz links using circular arrays
SummaryThis paper studies a multiuser downlink near-field sub-THz system for 6G networks, where the base station uses a large uniform circular array with a hybrid massive MIMO architecture. It minimizes transmit power under quality-of-service constraints using a two-stage penalty-driven joint optimization framework and a near-field polar-domain codebook. The hybrid precoder is obtained by solving a sparse approximation problem with a low-complexity SOMP algorithm. Simulations show performance close to the ideal fully digital beamfocusing scheme.
In briefEffect of Doppler Shift on Direct-to-Cell Satellite Links: Measurements and Proposed Improvements
SummaryThis paper examines direct-to-cell (DtC) satellites that serve unmodified LTE phones from low Earth orbit, where orbital motion causes Doppler shifts of tens of kHz. The satellite can pre-compensate Doppler for only one reference point per beam, so phones elsewhere in the beam retain a residual offset, which earlier field studies report can reach 10 kHz. The authors corroborate this with measurements of Starlink transmissions, link poor handover performance to inadequate residual compensation, and propose a UE-side method based on the UE's location relative to the beam reference point and the known satellite location.
Yuxi Chen, Muhammad Iqbal Rochman, Kyler McCormack et al.
In briefGARDEN: GPU-accelerated method for training billion-scale graph neural networks via co-designed caching, sampling, and batching
SummaryGARDEN is a GPU-accelerated system for training graph neural networks on billion-edge graphs. It co-designs caching, sampling, and batching to reduce costly CPU-to-GPU data transfers and inefficient memory access. On real-world billion-scale graphs, it reports end-to-end training speedups of 4.54x over DGL-UVA, 1.27x over DUCATI, and 2.97x over FastGL while maintaining convergence and accuracy.
In briefIterative interference cancellation for near-field XL-MIMO networks under line-of-sight channels
SummaryThis paper studies signal detection and interference cancellation for near-field users in extra-large MIMO systems under line-of-sight conditions. It proposes a weighted-rate successive interference cancellation (WR-SIC) scheme that orders users using the extra spatial degrees of freedom of near-field propagation, prioritizing those with stronger channels. It also develops an iterative detection and decoding (IDD) scheme with LDPC coding. Simulations with mixed near- and far-field deployments show improved bit-error and sum-rate performance over existing schemes.
In briefOptimal sampling and transmission for information freshness in gossip networks
SummaryThis paper studies how a transmitter should decide when to sample a source and when to transmit updates to several receivers that also exchange information among themselves. Freshness is measured with Version Age of Information (VAoI), and the problem is formulated as an infinite-horizon average-cost Markov decision process solved by relative value iteration. The authors prove that the sampling decision has a threshold structure in the transmitter's VAoI, and that under reliable links transmission and idling follow a VAoI-difference threshold structure. Numerical results show the optimal policy consistently outperforms natural baselines.
In briefQuantum synchronizable codes with polynomial and quasi-polynomial shifts
SummaryQuantum synchronizable codes protect quantum information against Pauli errors and block-boundary misalignments. The paper studies the case where the code's algebraic shift differs from the physical shift, analyzing constacyclic, polycyclic and quasi-polycyclic codes. It shows that the cost of misaligned positions can be corrected as erasures and that quasi-polycyclic codes over repeated-root trinomials achieve larger dimension and distance than known synchronizable codes of the same length.
Edgar Mart\'inez-Moro, Miguel de los R\'ios de Antonio
In briefEfficient SC-Flip decoding for polar codes using normalized decision thresholds
SummaryThis paper addresses the difficulty of identifying the first erroneous position in successive cancellation flip (SCF) decoding when only a few re-decoding attempts are allowed. It proposes decision-boundary normalization (DBN), which scales each decision log-likelihood ratio by the standard deviation predicted by Gaussian approximation, and a DBN-SCF decoding algorithm built on it. The method achieves better error-correction performance with limited re-decoding attempts and lower average decoding complexity than existing SCF methods at comparable performance.
Gulinisha Yiming, Zhongjun Yang, Zhicheng Liu et al.
In briefUpper limits on constant-composition codes via quantum channels
SummaryThis paper extends a quantum-channel optimization method of Alrabiah and Guruswami to q-ary constant-composition codes. It builds explicit channel and flag constructions that recover known classical asymptotic upper bounds, including the Ben-Haim–Litsyn, moving-subspace and Horn-channel bounds. The authors give Gram and flag formulations and show the bounds are stable under permutation transport and generalized shortening. Numerical optimization improves on known asymptotic upper bounds across all tested nonbinary parameter ranges.
Artur Akhiiarov, Peter Boyvalenkov, Danila Cherkashin et al.
In briefExtending DBLog incremental snapshots to remove orphan rows in CDC pipelines
SummaryThe paper addresses orphan rows that remain in a CDC sink but no longer exist at the source. It extends DBLog-based incremental snapshots so chunk reads can emit delete events for such keys, and adds a repair request that external processes can use to submit orphan candidates. Three modes are described, and the authors state that safety and convergence are formally verified in Isabelle/HOL and Lean 4.
In briefDICOMPUTE: A Framework for Cost- and Privacy-Aware Distributed AI Inference
SummaryThis paper proposes DICOMPUTE, a framework in which AI inference requests are scheduled under policy constraints across privately controlled, regional, and independently operated heterogeneous computing resources rather than only centralized facilities. It introduces a full-cost comparison covering capital amortization, energy, network transfer, and trust-related expenses, and a privacy model separating data residency, transport and execution confidentiality, and verifiable accounting. The authors conclude conditionally that distributed inference suits divisible, latency-tolerant or locality-sensitive workloads, and they provide falsifiable hypotheses and a reproducible experimental design.
Atul Khekade, Ritesh Kakkad, Anil Chinchawale et al.
