Papers

Data centers need tooling that validates an entire installation rather than individual nodes, at acceptance and at regular intervals thereafter. This requires dispatching identical benchmarks to every node in a single submission, and therefore cluster-aware scheduling. This paper presents ClusterBench, a framework for cluster-wide continuous benchmarking. It ships with a benchmark collection targeting each component: CPU, GPU, memory, interconnect, and I/O. Because measurements are repeated throughout the cluster's lifetime, ClusterBench collects data across space and time. Comparison against earlier runs detects performance regressions introduced by software changes, such as kernel updates or new library versions. The measurements also form a dataset for research on hardware variability. On the NHR@FAU clusters Helma, Alex, and Fritz, variation within a single component stays within 1%. Variation across specimens reaches 5%, despite nodes identical by specification. Correlating performance with power draw, frequency, and temperature shows that this relationship differs between air- and liquid-cooled nodes.

The Organic 6G vision of a network of networks spanning an edge-cloud continuum complemented by non-terrestrial resources requires, to realize its promise, service provisioning that is simple to operate, scalable across independently administered domains, and agile under domain churn (i.e., domains dynamically joining and leaving). Despite advances in cross-domain orchestration, many proposals rely on heavy integration fabrics, multi-layer coordinators, and deep telemetry pipelines that hinder deployability and amplify coordination overhead. We propose a lightweight, decentralized conversational orchestration framework based on Large Language Model (LLM)-driven domain agents. Each domain remains autonomous: an agent observes local state via tools, reasons in a closed loop, and exchanges summaries with neighboring agents over an Agent-to-Agent (A2A) overlay aligned with data-plane coupling. Fast feasible placement is enabled by periodic, routing-like dissemination of reachability advertisements (latency, bottleneck bandwidth, and compute capacity), while safe re-optimization, scaling, and migration are handled through event-driven requests and negotiation. To meet real-time constraints, we deploy a compact reasoning model trained with verifier-based self-verification and periodically refined online via shadow updates. Simulations show manageable, near-linear control-plane overhead as domains scale and during domain joins, and robust decision quality, including recovery after objective changes. We close by outlining future research directions for principled, secure, and uncertainty-aware agentic orchestration in Organic 6G.

Error-bounded lossy compression is essential for storing and transferring the vector-field data produced by large-scale scientific simulations. Although it enforces a user-specified error bound to limit numerical distortion, it does not preserve the field's topology: small admissible perturbations can create or eliminate critical points on which downstream feature analysis depends. Existing GPU compressors achieve high throughput but are topology-agnostic, whereas the only compressor with provable critical-point preservation (cpSZ) runs on the CPU at throughput far below the data-generation rates of modern GPU-based systems. We observe that, although preserving critical points is inherently a coupled and sequential constraint, it can be reformulated into independent parallel tasks, either on a per-block basis or, speculatively, on a per-point basis. We present FaCTz, the first GPU-based error-bounded lossy compressor that guarantees critical-point preservation. FaCTz provides a block-wise mode optimized for throughput and a speculative per-point mode optimized for compression ratio. Across three vector-field datasets, FaCTz preserves every critical point while achieving throughput of up to 60 GB/s, approximately two orders of magnitude (up to approximately 640x) faster than the multithreaded CPU implementation of cpSZ. Its speculative mode further improves the compression ratio by approximately a factor of two over the throughput-oriented mode.

In fault-tolerant quantum computing, fast and accurate decoding is crucial. Snowflake is a decoder for the surface code that runs in a streaming fashion. In this paper, we implement Snowflake on commercial FPGAs and validate them at room and cryogenic temperatures. Our results demonstrate high decoding throughput for small code distances that, when extrapolated, remains within acceptable limits for larger distances. Further, we incorporate the calculation of certain decoder confidence scores with negligible overhead both in terms of latency and physical resource utilisation. We note that implementing a large-scale system would require either a large FPGA beyond today's technology or clusters of FPGAs connected via a high-speed bus. Thus, we discuss an alternative architecture that exploits the locality of Snowflake by processing 2D slices of the 3D decoding window and offloading segments of the 3D structure to a high-speed memory.

Edge LLMs must preserve inference continuity when a user hands over between edge nodes, requiring key-value (KV) cache transfer to the target node. However, simultaneous handovers saturate the backhaul, preventing full cache delivery within the mobility-imposed transfer window. Rather than allocating bandwidth as if all cache entries were equally valuable, we order each user's KV cache by importance and transmit only its most informative fraction, turning token-level sparsity into communication savings. We cast the transfer as a multi-user backhaul allocation problem that maximizes average accuracy across users. Each user's partial-cache accuracy serves as its utility: a sigmoid that fits measurements on the RULER benchmark with $R^2>0.99$ across models and context lengths. Because importance ordering front-loads the high-value entries, the concave region of the accuracy curve spans nearly the entire cache. Our proposed allocator keeps served users within this region, making each per-slot allocation problem convex. The optimum is derived via a closed-form weighted water-filling solution that generalizes information-theoretic water-filling and enables online scheduling. The proposed allocator attains over 93.7% average accuracy in a 500ms transfer window, within 0.5pp of the full-cache ceiling, and reaches 98.2-99.5% of a clairvoyant upper bound.

Reinforcement learning (RL) for large language models is moving toward multi-turn agentic workloads, where rollout tasks repeatedly pause for external environments, resume with growing contexts, and finish at highly variable times. In this setting, RL training goodput, measured by training throughput, matters more than raw GPU occupancy: GPU waiting and repeated prefill recomputation are pure overhead. We present TideRL, a readiness-aware elastic RL system with Continuous Task Batching, Resource-Aware Ref-Actor Pipelining, and Elastic Resource Scaling. CTB preserves useful rollout state, $\textrm{RA}^2\textrm{P}$ selects between decoupled streaming and colocated aggregation from the ready backlog and arrival interval, and ERS moves ranks between rollout and training using the same readiness signals. Across text-only and multi-modal agentic workloads, TideRL improves RL training goodput by up to 5.6$\times$ over synchronous baselines and over 33% over asynchronous baselines, while reaching similar task performance. It also improves KV cache hit rate by 1.58$\times$, reduces per-step training time by up to 44.3%, and cuts total waiting time by up to 77.6%.

For an $n$-vertex graph of maximum degree $\Delta$, an equitable $(\Delta+1)$-coloring is a proper coloring all of whose color classes have size $\sigma = n/(\Delta+1)$ up to rounding. Known algorithms for relaxations of this target aggregate along a spanning structure, at cost scaling with the diameter $D$. We study near-equitable coloring in the LOCAL model and prove that, for deterministic algorithms, exact balance is inherently global: already on the cycle $C_n$, exact balanced free $3$-coloring requires $\Omega(n)$ rounds. More generally, additive imbalance $g$ requires $\Omega(n/g)$ rounds, matching a deterministic $O((n/g)\log^* n)$-round algorithm up to the $\log^* n$ factor, and these problems realize a dense family of intermediate deterministic LOCAL complexities $\tilde\Theta(n^\alpha)$, $\alpha \in (0,1)$, on cycles. The exact bound is sharp in two further senses: the identifier-universe threshold is exactly $N=n+1$, and the $\log^* n$ gap cannot be closed via faster ruling-set anchors. The bound then extends to twisted fiber products, an explicit family realizing, for every $\Delta \ge 3$, every diameter scale from the Moore bound $\Theta(\log n/\log\Delta)$ up to $\Theta(n/\Delta)$. Exact equitable $(\Delta+1)$-coloring requires $\Theta(D)$ rounds deterministically on this family, so exactness costs diameter time already on bounded-degree graphs of logarithmic diameter. In contrast, coarse balance admits local algorithms. On cycles, constant multiplicative equity costs $\Theta(\log^* n)$ rounds. On general graphs with large color classes, all class sizes can be kept within $(1\pm\eta)$ times the average in time independent of the diameter: with palette exactly $\Delta+1$ for $\Delta \le 2^{\sqrt{\log n}/C}$, and with palette $(1+\eta)(\Delta+1)$ for every $\Delta \le n^{1-o(1)}$, the latter in $O(\log n\,(\log\log n)^2)$ rounds.

The sequential part of GPU LZ77 decode is not where the field assumes it is. Across three decoder architectures on an H100 we measure that parse, not copy, holds 64-72% of device-resident decode time; that bounding back-reference chain depth - provable, and costing 0.006% in ratio - moves latency by at most 2.8% and, for the file's own latency spike, provably by nothing at all, since a byte-level comparison of all 15,499 blocks shows the cap alters none of the 181 blocks involved; that self-overlapping matches are periodic fills rather than dependency chains, which makes them fully parallel and speeds the match layer by 2.75-8.42x bit-perfect; and that the last genuinely sequential element, a four-entry distance history, can be removed by the encoder for 0.540% of ratio, growing the dependency-free parse run from 4 commands to 706. We also report the floor the format runs into: with a median match of 7 bytes against a 128-byte cache line, bus efficiency is 4.4% and a coalesced write of the same data is 39x faster. A separate section records ten hypotheses these measurements refuted, including one methodological error of our own. Every reproducible claim carries a machine-checkable record: a fresh clone of the tagged release passes 17 of 17 checks reachable without a GPU, none failing.

We consider federated parameter efficient fine-tuning of large neural networks with low-rank adaptation (LoRA,~Hu et al.\ 2022). Combining LoRA with federated PEFT introduces challenges absent from either setting alone: clients may use different LoRA ranks, making their factor matrices dimension-incompatible, and factor-wise averaging suffers from a bilinear mismatch. We propose SeFoRA, a sketch-aggregated federated LoRA algorithm in which each client transmits a linear sketch of its local updates, enabling direct aggregation at the federator. As a result, SeFoRA alleviates the bilinear mismatch, and allows for aggregation in a small subspace of the full model. We introduce a rank-homogeneous version called SeFoRA-Ho which allows for direct adapter aggregation in this setting. We prove convergence to a neighborhood of the first-order stationary point at rate $\cO(1/T)$ for the rank-homogeneous setting. Numerical experiments on fine-tuning RoBERTa-Large on GLUE datasets show how our algorithms outperform the state-of-the-art.

High-performance Tensor Core kernels rely on a low-level PTX pipeline built from asynchronous data movement with cp.async, warp-level matrix loads with ldmatrix, and matrix multiply-accumulate operations with mma.sync. However, most application code accesses Tensor Cores indirectly through the WMMA C++ API. This paper asks a focused, practical question: when does replacing WMMA with hand-written PTX actually pay off? To answer this question, we conduct a controlled, single-GPU study on an NVIDIA L4 GPU (Ada, SM89), comparing double-buffered WMMA baselines with a family of hand-written PTX GEMM kernels across FP16, INT8, and INT4 arithmetic and square problem sizes from $N=512$ to $N=8192$. Every kernel is profiled with Nsight Compute across the full metric set, and PTX speedups are reported relative to the corresponding same-precision WMMA baseline. Hand-written PTX provides no end-to-end speedup for FP16, because its instruction-level gains are offset by operand-packing overhead. In contrast, the PTX kernels achieve consistent speedups of 1.4x-1.8x for INT8, driven primarily by lower instruction counts and better global-memory coalescing, and 2.9x-4.3x for INT4, where native mma.sync.m16n8k64.s4 execution avoids the software-emulated sequence used by the WMMA path. Relative to the FP16 WMMA baseline, the best quantized kernels reach 34.4x (INT8) and 98.7x (INT4) at $N=8192$. Across these experiments, occupancy is a poor predictor of throughput. For large matrices, performance instead tracks memory-system behavior – particularly global-load coalescing and DRAM-active cycles – more closely than Tensor Core utilization. These results identify the precisions and operating regimes in which the additional complexity of hand-written PTX is justified.

Thread scaling in Hexaly on the Time-Dependent Vehicle Routing Problem with Time Windows (TDVRPTW) depends on modeling, and on the language binding through which the model reaches the solver. We compare two encodings, both reaching the solver through its C++ binding: one evaluates continuous travel-time functions exactly through external callbacks, which that binding evaluates concurrently, while the other approximates them with time slices evaluated natively by the solver. Independent CPU accounting confirms that both use the cores they are allocated, but they respond differently to width. The external-function encoding stabilizes, reducing its seed dispersion by about 30% up to 8 threads before widening again, yet its pooled quality barely moves, because nine of the ten instances improve while the hardest one degrades by enough to cancel them. For the time-sliced encoding we first select a discretization on feasibility under the original travel-time functions rather than on approximation error, and the retained setting then improves monotonically with the thread count, gaining about a quarter of its final gap and reducing the seed dispersion by nearly half between 1 and 16 threads. At equal single-threaded budget the two encodings are close, and the small pooled edge of exact evaluation comes from that same hardest instance, so the practical advantage of the discretization is thread scaling rather than fidelity traded for speed. These descriptive results support multi-threading as a way to improve solution quality and stability when the model can exploit it, and they show that binding-level constraints on external evaluation must be measured, not assumed.

A certified split point lets a parallel lexer cut unlexed input at a single byte with the serial token stream provably preserved, but several conventional token sets in the predecessor's controlled study certify no byte once string, comment, or whitespace-run forms are included (arXiv:2608.03473). We generalize from a byte to a bounded window: a byte string after which the position where the current token began is known, regardless of surrounding context. We certify the directly usable form of that recovery: the token covering the window's final byte begins at the reported origin. The certificate is conditional on occurrence and may be vacuous; every applicability figure counts only windows carrying an asserted completely tokenizable occurrence witness. We give a conservative model of a maximal-munch scanner's possible histories across a window, prove it sound, and decide reachability in that model exactly by exhausting a finite quotient of its reachable configurations, so every answer of the unbudgeted procedure is either a certified window with its origin or a proof that the model admits none. Within the stated flat, completely-tokenizable scope, model-positive answers are semantic certificates; negatives are relative to the conservative model, which deliberately refuses some windows a greedy scanner would allow. A token set in which some token matches the empty string is decided through its positive-width equivalent, the same automaton entered through a start state that does not accept, which changes no scan. In a sample of 400 random token sets, 322 of the 337 sets certifying no byte gain a witnessed window, with zero inconclusive searches, and every exact-empty row of the predecessor's study gains a witnessed window of two to four bytes. The analysis runs once after automaton construction, using only the compiled tables and no input.

Decentralization as a concept in computer science has existed for over half a century. Despite its fundamental role across domains such as security, distributed computing, artificial intelligence, cloud infrastructures, and Internet of Things (IoT) architectures, there remains no universally accepted definition of decentralization applicable across computer communication systems. This has become increasingly problematic with the emergence of decentralized AI and machine learning paradigms, including collaborative training, distributed inference, blockchain-based, and agentic AI, where decentralization is often treated as a core design objective. Meanwhile, existing approaches frequently conflate decentralization with related notions such as distribution of trust or specific implementation paradigms. Such ambiguity creates inconsistencies in system analysis, limits comparability between works, and weakens the rigor of formal reasoning surrounding communication architectures and protocol design. In this work, we define this research gap as the Decentralization Problem. We analyze the formal-semantic, epistemological, and pragmatic foundations of decentralization and introduce a graph-based ontology defining it as both relational and subject-specific property of computer communication systems. The framework formally distinguishes decentralization from distribution and supports evaluation through two novel metrics: Void Tolerance and Imperviousness. We also provide a browser-based implementation that enables automated classification and metric computation of arbitrary systems. Instantiations to federated learning and blockchain architectures show consistent, comparable assessments where existing definitions produce incomplete or contradictory conclusions, providing a domain-independent foundation for analysing decentralization across heterogeneous systems.

Low-rank adaptation (LoRA) represents large language model (LLM) updates with two compact matrix factors, i.e., $A$ and $B$, providing an efficient way to fine-tune large models in federated learning paradigm. Inspired by the asymmetric roles of the LoRA factors, we study whether $A$ should be shared across clients while $B$ remains client-specific (Share-A/Local-B), or whether $B$ should instead be shared while $A$ remains client-specific (Share-B/Local-A). With a least-squares surrogate, we reveal that Share-A/Local-B requires the client-specific LoRA update matrices to use a common rank-$r$ input-side space, whereas Share-B/Local-A requires a common rank-$r$ output-side space. The two strategies therefore incur different projection residuals, indicating that the preferred strategy is the one with the smaller aggregate residual across clients. With this insight, we propose Federated Adaptive Factor Sharing Low-Rank Adaptation (FedAS-LoRA), which selects the sharing side before training to enhance fine-tuning performance. To enable adaptive factor selection before training, we design a Rank-Aware Shared-Subspace Sufficiency (RSS) metric, which effectively assesses whether a shared rank-$r$ input subspace is sufficient for the local data distributions using representations extracted from a frozen LLM backbone. Experiments across different tasks, data distributions, LoRA ranks, and participation settings confirm the effectiveness of RSS and the superior performance of FedAS-LoRA.

The symmetric rank-$k$ update (SYRK), $C = AA^\top$, computes the dot product of each pair of rows of $A$, producing the Gram matrix $C$. Its sparse variant underpins similarity search in machine learning, graph analytics, and genomics, including Jaccard similarity on datasets too large for a single node. Despite the symmetry in its inputs and outputs, existing distributed sparse matrix multiplication algorithms such as Sparse SUMMA treat sparse SYRK as generic multiplication, computing the full output and materializing the explicit transpose even when the calling application uses only one triangle. Prior distributed $AA^\top$ computations in similarity search and genome assembly inherit this overhead from the underlying SpGEMM. This paper presents SpSYRK and CommSpSYRK, two distributed sparse SYRK algorithms that exploit symmetry. The first, SpSYRK, partitions the off-diagonal blocks of the output between the upper and lower triangular regions of the process grid and computes only the lower-triangular part of each diagonal block, halving per-process computation compared with state-of-the-art distributed SpGEMM. The second, CommSpSYRK, further reorders communication to avoid forming $A^\top$, which reduces per-process communication volume. On 32 nodes of the Perlmutter supercomputer, SpSYRK achieves a 2$\times$ speedup over an optimized Sparse SUMMA on matrices where local multiplication dominates the runtime; the advantage narrows on communication-bound inputs, a dependence that the cost model predicts from the arithmetic intensity. CommSpSYRK fixes this and consistently achieves superior scaling at high process counts. The approach is a drop-in replacement for any application computing $C = AA^\top$ via a distributed SpGEMM routine, and its triangular output can be consumed directly by subsequent operations, reducing both computation and memory footprint.

CPU-GPU coscheduling enables simultaneous execution of an application across both processing units, but its efficiency depends on workload partitioning and memory architecture. This preliminary study evaluates coscheduling on the NVIDIA GH200 Superchip compared to a discrete H100 PCIe platform. Using sparse conjugate gradient (CG) as a case study, we assess various work divisions across three memory-management paradigms: explicit copy, managed memory, and mapped memory. Our evaluation highlights the run time and programmability tradeoffs of reducing manual CPU-GPU data movement. The results show that compared with the H100 PCIe platform, GH200 makes several hybrid CPU-GPU work divisions competitive and makes managed memory practical for several matrices. These results suggest that integrated CPU-GPU platforms such as GH200 can improve both performance and programmability for coscheduled workloads.

Persistent efforts in recent years have been devoted to devising distributed algorithms for fundamental optimization problems in planar graphs. In particular, for Single-Source Shortest-Paths, there is an $\tilde O(D^2)$-rounds exact algorithm [Li, Parter STOC'19] for directed planar graphs, and an $\tilde {O}(D)$-rounds $(1+o(1))$-approximation algorithm [Rozhon, Grunau, Haeupler, Zuzic, Li STOC'22] for undirected planar graphs (where $D$ is the graph's hop-diameter). Recently [Abd-Elhaleem, Dory, Parter, Weimann PODC'25], a matching bound for the exact case was obtained for the Maximum $st$-Flow problem. Namely, an $\tilde O(D^2)$-rounds exact algorithm for directed planar graphs. However, for the approximate case, they give a $D\cdot n^{o(1)}$-rounds $(1-o(1))$-approximation algorithm for undirected planar graphs that works only for the special case where both $s$ and $t$ lie on the same face. In this paper, we remove the restriction that both $s$ and $t$ must lie on the same face (we also eliminate the $n^{o(1)}$ factor). Namely, we present the first distributed near-optimal $\tilde{O}(D)$-rounds $(1-o(1))$-approximation algorithm for Maximum $st$-Flow in general undirected planar graphs. Our main technical contribution is a distributed implementation of the classical Reif's [SICOMP'83] centralized algorithm. This is achieved by a careful recursive incision procedure on the planar dual $G^*$ of the graph $G$. It is challenging, because we need to simulate dynamic changes (incisions) over the dual graph $G^*$, while we can only communicate over the input graph $G$.

A main promise of looped language models is depth-adaptive inference. By looping a block of shared layers a variable number of times, the model can use less compute for "easy" tokens and more for "hard" ones. However, tokens with different numbers of loops cannot share a uniform forward pass and therefore cannot be handled by standard batching systems such as vLLM. The practical value of depth-adaptive inference thus hinges on whether batching can be made efficient. We introduce the first efficient method for depth-adaptive looped LMs via continuous depth batching (CDB), which forms new batches between loop steps. Our method dynamically schedules looped and non-looped parts of the architecture, manages looped KV-caching, and predicts which tokens will exit the loop in advance so it can prepare batches asynchronously. Experiments on Ouro 1.4B and Huginn 3.5B show that fully looped architectures are best suited to depth-adaptive inference, as large non-looped layers outside the recurrent core (e.g., token embedding, LM head, and unshared transformer blocks) slow down and complicate scheduling. Overall, CDB realizes up to 99% of the estimated maximum speedup available, leaving further gains primarily dependent on model architecture and exit behavior.

Distributed batch data processing applications are widely executed on cloud-based resources where restricted user access to node-level hardware energy counters hinders transparent sustainability accounting. Energy and carbon attribution methodologies therefore depend on power models and available resource utilisation traces, yet the accuracy of these estimates has to be validated while direct counters are available. In this work, we use Apache Spark running on Kubernetes as a case-study dataflow runtime and cluster resource manager to compare model-based energy estimates to Intel RAPL package and DRAM energy on an AWS bare-metal cloud and an on-premises cluster, comparing different CPU usage signals and memory coefficients. We show that external monitoring improves signed package-energy error relative to Spark task traces, reducing underestimation from -29.58% to -24.41% on AWS and from -24.00% to -16.22% on-premises.

Workload scheduling in cloud environments often relies on simplistic assumptions about application resource needs and hardware utilization. Overlooking application-level performance objectives and hardware resource contention that leads to inefficient resource usage and degraded performance. This paper addresses two key limitations of current approaches. First, unnecessarily strict enforcement of service level objectives (SLOs) often leads to resource underutilization and poor energy efficiency. Second, lack of congestion awareness in shared resources such as last-level cache (LLC) and memory bandwidth. In this paper, we propose two complementary strategies to address these limitations: (i) integrating soft SLO limits that allow controlled overcommitment and tolerate minor, transient violations to improve cluster efficiency, and (ii) introducing resource-aware scheduling and rescheduling based on real-time congestion insights for shared resources such as last-level cache (LLC) and memory bandwidth. Our results show that soft SLO limits reduce corrective rescheduling actions by 49% compared to hard-limit approaches while maintaining acceptable performance guarantees. Additionally, resource-aware scheduling decreases node-level congestion by 8% and further mitigates SLO violations, demonstrating the effectiveness of incorporating application-level flexibility and hardware-level insights into scheduling and rescheduling decisions.

Edge LLM inference combines sparsity and low-bit quantization to meet device memory, latency, and power limits. Yet quantization shrinks weight payloads without proportionally reducing sparse metadata, so index traffic and nonzero extraction become critical SpMM bottlenecks. We introduce the Payload-to-Metadata Ratio (PMR) and show that improving PMR raises effective compute intensity in decoding. We present UnionSparse, an index-efficient framework that combines Index-Efficient Bitmap Encoding (IE-BME) with a SpMM kernel using Low-Bit Shared-Memory Parallel Decoding (LSPD). IE-BME amortizes metadata and aligns sparse traversal with fragment assembly, while LSPD improves small-batch execution. Under W4A4 quantization and 30%–70% sparsity, UnionSparse outperforms FlashLLM and SpInfer by 2.30x and 1.43x, and CUTLASS and cuBLAS Tensor Core by 1.56x and 3.46x, respectively. These results establish payload-extraction efficiency as a first-order concern for low-bit sparse inference on edge GPUs. Source code is available at: https://github.com/Victor-Alen/UnionSparse.

Federated learning (FL) must serve devices with varying computational capabilities. A fixed model cannot suit all devices, while training one model per deployment limit is costly. Federated supernet training instead learns one elastic model with differently sized subnetworks, then deploys a suitable one to each device. When client inference budgets differ, however, parameters exclusive to high-cost subnetworks are reachable by fewer clients. We propose FEAST, a federated shared-space training framework that counters this imbalance by jointly training multiple subnetworks within each client's limit. Budget-tailored sub-supernet routing sends only the relevant supernet portion, and sparse aggregation merges the returned parameter slices. The trained supernet directly serves the subnetworks used during federation and supports post-hoc extraction of additional subnetworks without federated retraining. We further show that independently assigning clients' training-data volumes and inference budgets can distort accuracy–inference-cost comparisons in heterogeneous FL simulations, and introduce a one-parameter $\gamma$-allocation protocol to control this coupling. In our experimental setup, the SuperFedNAS and DeepFedNAS supernet training procedures remain near chance at 25M and reach at most $17.09\%$ at $596$M inference MACs; FEAST reaches $71.06\%$ at $596$M, $2.4$ points above the strongest model-heterogeneous weight-sharing baseline at its largest tier. Across CIFAR-100, CINIC-10, and TinyImageNet-200, FEAST achieves the highest population-averaged accuracy among the evaluated weight-sharing methods when each client receives its largest affordable subnetwork. Sub-supernet routing reduces aggregate model-parameter traffic by $6.8\times$ relative to full-supernet transmission.

Privacy-preserving payment systems are well understood, yet concerns about their misuse for financial crime have led to only limited adoption in regulated settings such as central bank digital currencies (CBDCs) and institutional stablecoins. Tracing is one tool for addressing these concerns: acting on external evidence implicating a user, law enforcement follows the suspect's funds through the ledger to uncover laundering routes and accomplices. Existing coin-tracing schemes, however, provide no cryptographic bound on tracing reach: once initiated, a trace may propagate indefinitely through the transaction graph or persist across all future transactions of a targeted user. Keeping surveillance targeted and temporary therefore depends on the restraint of the authority or a committee. We introduce ephemeral coin tracing (ECT), a primitive that bounds tracing reach by construction. Each account carries an encrypted tag that records which traced identifiers its funds carry while hiding its tracing status from users. When funds move, the sender's tag degrades and merges with the recipient's tag. Each tracing contribution expires independently after a policy-defined number of hops and then becomes unrecoverable even to the tracing authority, without affecting other live contributions in the same tag. Public parameters also bound how many identifiers a tag can distinguish simultaneously. We formalize ECT and give constructions based on exponential ElGamal, Damgård-Jurik encryption, and Ring-LWE, the last providing post-quantum security.

Federated Learning (FL) enables collaborative model training across distributed client devices while preserving data privacy. However, FL faces significant challenges due to data heterogeneity, particularly in terms of label distribution skewness and variations in dataset sizes, which can lead to biased model updates and hinder convergence. To address this, we propose FedTVD, a novel FL algorithm that weights client contributions during aggregation by considering both data quality and quantity. Unlike traditional FL approaches such as FedAvg, which rely solely on dataset size for client weighting, FedTVD integrates Total Variation Distance (TVD) to measure the divergence between each client's local label distribution and a uniform global distribution. Clients with highly skewed distributions receive lower weights, preventing unbalanced datasets with imbalances from disproportionately influencing the global model. At the same time, dataset size is incorporated to ensure scalability and fairness. This dual-weighting mechanism effectively mitigates the impact of data imbalance, leading to more stable and generalized global models. Experimental results show that FedTVD consistently outperforms state-of-the-art methods across all datasets (FMNIST, CIFAR-10, and CIFAR-100) and all levels of data heterogeneity. Notably, it achieves up to 10.6% improvement over FedAvg on CIFAR-10 under highly skewed data, while maintaining top performance even under moderate and IID settings.

Decentralized intelligence systems with heterogeneous devices and limited coordination increasingly rely on decentralized federated learning (DFL). However, DFL suffers from convergence inefficiency under data heterogeneity due to the use of a uniform learning rate (LR) that ignores layer-specific optimization needs. Foundational layers are responsible for maintaining network consensus, while specialized layers adapt to local data characteristics, leading to conflicting gradients and degraded performance under non-IID conditions. To address this fundamental tension, this work introduces FedA2L, a method that dynamically adjusts layer-wise LRs based on model divergence signals. By leveraging local update intensity and network consensus constraints, FedA2L seamlessly integrates into existing DFL protocols without additional communication or coordination. Extensive evaluations across DFL algorithms, various model architectures, and datasets demonstrate that FedA2L achieves up to 4.94 times faster convergence than vanilla DFL and reduces communication rounds by up to 59% compared to scheduler-based baselines. Furthermore, FedA2L exhibits resilience to severe data heterogeneity, larger network sizes, and sparse topologies, reducing communication overhead and establishing it as a versatile optimization tool for resource-constrained or large-scale distributed learning in edge and IoT deployments. The code is released at https://github.com/nclabteam/FedA2L.

Modern processors increasingly provide matrix engines whose peak arithmetic throughput greatly exceeds conventional SIMD, but scientific applications rarely realize this advantage end to end. We examine this gap in SPECFEM3D's dominant stiffness operator on the Arm LX2 CPUs that power the flagship Lineshine supercomputer. Against a matched, high-performance SVE baseline on the same cores, SME's $4\times$ single-precision peak advantage falls to $2.2\times$ for isolated tensor contractions and $1.1\times$ for the complete operator. Our factorized diagnostic attributes the loss to pointwise computation, indirect field movement and synchronization, and irregular coefficient delivery. Explicit SIMD mitigates pointwise work, raising the full-operator speedup to $1.3\times$. Field-layout changes mitigate indirect movement and synchronization, while vector-blocked coefficient streaming reduces irregular-access costs; together they raise speedup to $1.6\times$ at high order. A contraction-free control bounds further contraction-only gains at $1.11$–$1.32\times$. Realizing matrix-engine performance therefore requires co-designing the entire operator path, not merely replacing its contraction kernel.

Large-scale managed cloud databases leverage sophisticated load Packing and Migration (PAM) algorithms, which provide the efficiencies necessary for running these services at scale on cloud resources. Research into optimizing the resources and reliability of cloud databases at massive scales is limited by a lack of public NoSQL workloads. We address this in the context of Cosmos DB, Microsoft's flagship cloud-hosted NoSQL database. We first propose open-source NoSQL workloads from real Cosmos DB clusters, and analyze these traces to derive a novel reliability metric, Distressed Resource Volume (DRV), which captures the quality of service experienced by the end user. We then develop an open-source policy simulation framework, LoadStar, powered by a non-parametric statistical model of estimating the QoS of real traffic patterns. These form a reusable benchmark pipeline for validating policies for resource-centric NoSQL workloads. We then define a resource optimization problem for placing Cosmos DB replicas onto VM nodes, develop the Luna model for forecasting future load distributions, and the Orbit PAM algorithm that uses these forecasts to trigger and rebalance stressed replicas, to reduce tail-errors. Our experiments, validated using LoadStar for these workloads, demonstrate Orbit's benefits over the existing Cosmos DB policy and a worst-fit optimized baseline, with higher load delivered at lower error rates and up to $35\%$ reduction in resources. These have been deployed in production, with potential savings of $\$100M$s/yr while improving service reliability for millions of customers.

Query-Key Normalization (QK-Norm) improves the training stability and quality of modern Large Language Models (LLMs). However, under Tensor Parallelism (TP), layerwise QK-Norm introduces additional cross-GPU communication because the normalization factor depends on the full hidden vector. We present SwiftQK, a multi-GPU RMSNorm kernel that exchanges only scalar normalization statistics and overlaps the remaining Peer-to-Peer reduction with independent element-wise computation in a deadlock-safe persistent kernel. Evaluations on recent LLMs show that SwiftQK reduces QK-Norm latency by 81.4–93.9% relative to the standard TP QK-Norm using full-vector All-Gather. In end-to-end serving, SwiftQK reduces TPOT on average by 29.5% over the All-Gather-based baseline and by 14.3% over an optimized scalar-aggregation implementation.

We present PDHCG-CQP, a GPU-accelerated first-order solver for large-scale conic convex quadratic programming. PDHCG-CQP supports affine constraints and Cartesian products of nonnegative, second-order, rotated second-order, exponential, and three-dimensional power cones. At its core is a restarted averaged primal-dual hybrid gradient (PDHG) method, whose primal update is computed inexactly by solving a conic quadratic proximal subproblem with projected gradient iterations. We establish local linear convergence of the restarted averaged scheme with both exact and inexact primal proximal evaluations under a uniform local quadratic-growth condition on the smoothed primal-dual gap. We further show that this condition holds under strict complementarity by exploiting a rotated second-order-cone lifting together with local primal and dual regularity conditions. Our C/CUDA implementation combines matrix-free linear algebra, batched cone projections, adaptive inner solves, reflected-Halpern acceleration, and fully device-resident KKT residual computations. It also supports multi-GPU execution through a two-dimensional partitioning of the problem data. Extensive experiments on standard and large-scale quadratic programming (QP), convex quadratically constrained quadratic programming (QCQP), second-order cone programming (SOCP), and quasilinear Fisher equilibrium benchmarks demonstrate that PDHCG-CQP achieves state-of-the-art robustness among first-order solvers while scaling efficiently to 8 GPUs and instances with up to $4.4\times10^8$ stored primal coordinates. PDHCG-CQP is open source and available at https://github.com/Lhongpei/PDHCG.

Deterministic rendezvous for two anonymous mobile agents starting simultaneously from two distinct nodes of an anonymous connected graph and navigating synchronously in the graph requires that they meet at some node. An instance of the rendezvous problem is the underlying graph, together with two distinct nodes that are the initial positions of the agents. Such an instance is said to be feasible if there is a deterministic algorithm guaranteeing rendezvous, possibly valid only for this instance. A rendezvous algorithm is said to be universal for a class of instances if it guarantees rendezvous for all feasible instances from this class. We consider the model with footprints: whenever an agent visits an unmarked node, it leaves a permanent footprint on it, and all footprints are identical. This paper aims at answering the open problem from the paper by Das and Pelc (SPAA 2026), asking whether there exists a universal rendezvous algorithm for the class of all instances in the model with footprints. We propose a universal rendezvous algorithm for the class of all instances where the underlying graph is connected (finite or countably infinite) in the model with footprints. This shows the existence of a universal algorithm for the class of all instances in the model with footprints, which is an affirmative answer to the open problem.