Papers

Exascale simulations generate data far faster than it can be stored or analyzed, making efficient data reduction essential. Error-controlled lossy compression offers high compression ratios under user-specified error bounds, but the target tolerance must be fixed at compression time. Progressive compression relaxes this restriction, yet existing methods still rely on fixed refactoring strategies and do not fully exploit correlations among decomposed coefficients, limiting the efficiency of progressive retrieval. In this work, we present an adaptive progressive compression framework that improves retrieval efficiency for two common targets, namely error-bound and peak Signal-to-Noise ratios. Our contributions are fourfold. (1) We propose to leverage two complementary interpolation schemes for adaptive progressive compression toward different targets, and we optimize them to achieve high efficiency. (2) We propose coefficient decomposition, a novel method that exploits the commonly overlooked spatial correlations among decorrelated data, which further improves the efficiency. (3) We develop the adaptive progressive compression workflow with automatic selection of the best-fit refactoring pipeline and tailored optimizations. (4) We evaluate the proposed framework on five real-world scientific datasets against three state-of-the-art progressive compressors. Experimental results demonstrate that the proposed framework improves the compression ratio by up to $42.3\%$ under the same requested error tolerance and up to $92.5\%$ at the same PSNR, compared with the best-performing existing methods. When transferring $512$ GB of scientific data to remote sites, the framework delivers up to $1.26\times$ speedup in the end-to-end data transfer performance. Furthermore, our method achieves the highest visualization quality while retrieving the least amount of data from storage.

Large language model (LLM) sustainability is increasingly a serving-systems problem, not only a training problem. In production, energy and carbon impact depend on more than model size: workload shape, batching, key-value (KV) cache reuse, prefill/decode placement, model and accelerator choice, power state, geographic carbon intensity, and service-level objectives (SLOs) all matter. Recent systems papers study many of these factors separately. This paper connects those results and asks a practical engineering question: what do they imply when the decision point is a distributed inference control plane such as llm-d? The contribution here is synthesis, not a new set of benchmark results. Reported performance, energy, carbon, and cost improvements remain the results of the cited papers and systems. I group the literature into recurring design patterns and use those patterns to sketch a Sustainable Inference Control Plane (SICP) for llm-d. The proposed control plane would consider latency, energy, carbon, cache reuse, serving cost, and quality when routing and scaling, while keeping TTFT/TPOT SLOs as hard constraints. I also outline an evaluation framework based on SLO-satisfied goodput per joule and per gram CO2e, together with a reproducible experimental plan. The main observation from connecting the literature is that sustainable LLM inference is unlikely to come from one "green" model or one accelerator; it is more naturally treated as a control problem across model, phase, cache, hardware, replica, region, and time.

As an increasing number of embedded platforms incorporate multiple processing units, shared resource contention induced unpredictable execution time poses a challenge for real-time system design. Memory bandwidth regulation is a popular mitigation approach, and MemGuard is the canonical example. Recently, MemPol introduced a new bandwidth regulation mechanism, which was compared with MemGuard. Specifically, they reported significant overheads for MemGuard, citing up to a 1.79x slowdown, to contextualize MemPol's comparative benefits. This report is meant to clarify and add the necessary nuance to the experiments carried out in that prior work. Specifically, we show that the MemGuard overheads presented in these prior evaluations were unintentionally amplified as the result of using a suboptimal configuration with an older version of MemGuard, wherein the benchmark under test was pinned directly to the master core responsible for handling global timer interrupts. By faithfully reproducing these specific experiments using a modern, decentralized implementation of MemGuard, we demonstrate that the actual execution overhead drops significantly under identical conditions. Consequently, when evaluated with a properly configured recent version, MemGuard exhibits an overhead that is highly comparable to MemPol's overhead. By revisiting these baseline metrics, this report provides an updated and comprehensive perspective required for future evaluations of memory bandwidth regulators.

Distributed quantum computing (DQC) offers a promising approach to scale quantum computing by overcoming the resource limitations of a single quantum processor. However, inter-node communication remains a major bottleneck of DQC due to inefficient and error-prone entanglement distribution. Optimizing inter-node communication can not only reduce the amount of entanglement resource needed to execute a quantum circuit but also improve execution speed and accuracy of the results. This paper proposes DPRQ, a qubit routing algorithm for minimizing inter-node communication in distributed quantum circuits divided into collective communication blocks. Unlike current approaches that utilize greedy block-level qubit routing strategies, DPRQ employs a dynamic programming-based technique focused on global circuit-level optimization, while capturing inter-block dependencies. We evaluated DPRQ on four sets of quantum circuits and a variety of DQC configurations. The results demonstrate that DPRQ's innovative routing strategy achieves an average of 24.40% reduction with a maximum of 85.06% reduction in inter-node communication, when compared to the state-of-the-art collective communication-based DQC compiler QuComm.

Compound AI workflows are increasingly used to serve complex AI tasks by coordinating multiple AI models and software components. This approach enables deployment flexibility, as each workflow stage can expose different model variants and resource requirements, but it also expands the deployment choices. A deployment must choose an execution plan that selects AI models for each compound AI workflow stage and places them on a heterogeneous cluster in order to satisfy SLOs. Deployment optimizers therefore need estimates to compare many candidate plans and identify feasible ones. System metrics can often be profiled per stage and composed according to workflow topology, but accuracy cannot, as errors and information loss at upstream stages affect the accuracy of downstream stages. Existing approaches either profile complete configurations end to end, which scales poorly, or use product-based accuracy surrogates that treat stages as independent and can misrank candidate plans. We introduce Atlas, a framework for optimizing compound AI deployments under SLO constraints. Atlas uses MAP, a Markovian Accuracy Predictor, to estimate configuration accuracy from local conditional accuracy transitions between adjacent workflow stages. MAP discretizes intermediate outputs into accuracy buckets and composes transition profiles according to workflow topology, giving the optimizer an accuracy estimate without exhaustive end-to-end profiling. Atlas formulates execution-plan selection as a mixed-integer linear program that maximizes predicted accuracy subject to SLOs. Across four compound AI workflows, MAP achieves Spearman correlation up to 0.947 while reducing profiling cost by up to 2.6x relative to exhaustive end-to-end profiling. Guided by MAP, the Atlas optimizer selects execution plans within 0.03 of oracle accuracy while reducing deployment cost by up to 42% through heterogeneous placement.

Confidential serverless computing is rapidly emerg- ing as a critical paradigm for application domains requiring strong confidentiality guarantees, such as healthcare, finance, and machine learning. To enable this paradigm in untrusted cloud environments, Confidential Virtual Machines (CVMs) provide isolation by encrypting the entire guest memory, and thus securing serverless workloads against host-level access and inter- ference. However, the implications of CVMs for serverless systems remain insufficiently understood. This paper presents an empirical study of serverless work- loads in CVMs, systematically covering memory efficiency and runtime overhead on both warm-starts and cold-starts. Our results show that CVMs incur substantial memory overhead because encrypted memory disables cross-VM page deduplication and reduces memory reclaimability, thereby limiting warm- container capacity under a fixed memory budget. Runtime overhead in warm-starts is workload dependent. Workloads with frequent VMEXITs, particularly idle transitions, suffer substantial slowdowns. These slowdowns are further amplified by common serverless deployment practice that couples vCPU allocation to memory size, as higher-memory configurations expose more vCPUs than some functions can use effectively. For cold starts, the study focuses on container creation, a common and major contributor to startup latency. Motivated by the memory-efficiency results, we consider the deployments in which multiple containers of the same function are consolidated within the same CVM to improve efficiency, and show that selectively relaxing certain isolation mechanisms in this setting can substan- tially reduce startup overhead. These results clarify the main performance tradeoffs of confidential serverless computing and suggest practical ways to improve efficiency and latency.

Distributed AI training involves recurring rounds of data exchange between multiple pairs of GPU nodes. Slowdown in even one flow due to congestion can cause the entire communication round to slowdown. Current approaches for evading congestion in AI clusters assume global control over the entire workload (e.g. coordinating the schedule of all jobs) or assume infrastructural support (e.g. adaptive routing in switches). They are thus ill-suited in a shared cloud setting where AI jobs belonging to one user can face external congestion from other users' jobs or background traffic beyond its own control. In this paper, we build a system, REACT, that tunes the recurring pattern of data exchange between GPU nodes (known as communication collectives) in response to congestion. REACT works at the application (communication library) layer, where it detects congestion at runtime using readily available flow stats, and tunes the collective pattern to alleviate congestion - changing the set of incident flows while retaining the semantics of information exchange (e.g. selecting which node aggregates data in an AllReduce tree). REACT requires no explicit support from the underlying network infrastructure and can be unilaterally deployed by individual users in a shared cloud setting. We prototype REACT as a shim layer over NCCL, and evaluate it on a shared academic GPU cluster - enabling REACT improves communication performance (algorithm bandwidth) by 13%-38% under network congestion. Our simulations across a range of congestion scenarios further reveal up to 75% performance improvement, highlighting the effectiveness of our approach.

Modern atomistic simulations increasingly employ localized orbitals to represent quantum operators, yielding sparse block matrices whose block shapes vary with chemical species and basis choice. Conventional scalar sparse formats store the entries of each block individually, obscuring this local structure and limiting the use of efficient block algorithms. We present VBCSR, a distributed sparse matrix library that preserves variable-size atomic blocks and accelerates the core linear algebra of large-scale atomistic simulations. A unified interface automatically maps scalar, uniform-basis, and multispecies operators to compressed sparse row (CSR), block sparse row (BSR), or variable-block compressed sparse row (VBCSR). Our advanced acceleration method groups blocks of equal shape and dispatches them to optimized dense kernels. In the reported benchmarks, VBCSR outperforms the tested Python-accessible reference implementations for several block-sparse benchmarks. We further demonstrate VBCSR in an InP nanoparticle application containing more than \(10^6\) atoms.

We present a systems-security case study of a two-node split-LLM training system whose privacy evaluation passed while leaving an observable channel untested. The Trusted Local Node (TLN) sends protected activations to the Untrusted Cloud Node (UCN), the UCN returns its output, and TLN, holding the private loss, returns the output gradient. The frame the UCN receives mixes real rows with decoys, and the loss ignores the decoys. Their gradients are exactly zero, so the pattern of zeros reveals which rows were real. We measure it with a protocol fixed in advance: a leak injected at known strength to prove the instrument can see one, a shuffled-label control to prove it does not report absent leaks, and a threshold set before the runs. Across nine seeds, the zeros identified the real rows on every frame, 4,096 of 4,096 per run. An attack on the frame contents recovered about one extra token per hundred over a constant-guess baseline (+0.65 to +1.50 percentage points); the shuffled controls recovered nothing. A second set of runs repeated this on a configuration that keeps model quality within budget, so the finding is not confined to a setting nobody would deploy. On both datasets, every such run passed the forward-channel privacy check and the quality check, yet failed that same check once the returned gradient was included. Clipping and noising each row of the gradient closed the leak for about 0.01 nats of held-out cross-entropy. The system is not thereby safe: five classes of attack, including those accumulating observations across training steps, were never measured.

As edge-based deep learning applications become more complex, optimizing performance on heterogeneous System-on-Chips (SoCs) presents unique challenges. Traditional pipelining techniques distributing the computation across different on-chip processing units, while effective for throughput, do not address the latency demands posed by modern neural networks with complex interdependencies and extensive operator parallelism. There is a potential in leveraging operator parallelism to enable concurrent execution across multiple processing units, thereby reducing inference latency. However, prioritizing pipelining or parallel execution often necessitates a compromise, where optimizing one performance metric adversely impacts the other. This paper introduces Para-Pipe, a hierarchical mapping framework that integrates intra- and inter-stage operator parallelism within a pipelined architecture. Para-Pipe navigates the trade-off between throughput and latency by selectively fine-tuning parallelism levels within and across pipeline stages. This strategy can significantly reduce inter-processor communication overhead, significantly improving energy efficiency. Our evaluation demonstrates that Para-Pipe generates multiple Pareto-optimal configurations, achieving a balance between throughput and latency on an Amlogic SoC equipped with ARM big.LITTLE CPUs and GPU, as well as the Black Sesame Technology SoC featuring a deep learning accelerator and two DSPs. More importantly, throughput-optimized configurations under Para-Pipe on Amlogic SoC show an average energy efficiency improvement of 11.0% over purely pipelined strategies and 23.3% relative to non-pipelined parallel execution.

In DAG-based consensus, all validators propose blocks concurrently, and designated leader blocks drive transaction commit. Having multiple leader slots per round cuts queuing latency, yet production deployments run a single leader because of head-of-line blocking: a slow leader stalls the pipeline for at least one leader timeout, and for several waves when its slot must wait for the fallback indirect decision rule. This risk grows with the leader count. We introduce Barnacle, an add-on that adapts the leader count at run time. Every interval, it measures on the agreed committed DAG the fraction of slots decided as commit by the direct rule, and drives the leader count with additive increase, multiplicative decrease. The measurement requires no extra messages and no cryptography, and is deterministic. Barnacle is generic over DAG protocols; we instantiate it on four protocols spanning the Byzantine (3f + 1, 5f + 1), crash-only (2c + 1), and mixed (5f + 3c + 1) fault models, with proven safety and liveness. Results show Barnacle matches the best static leader count in every regime: in a healthy network its latency is 6-13% lower than a single leader's, and under degradation it matches a single leader while remaining 35-56% below a static high count. We are currently collaborating with the Sui team to integrate Barnacle into the Sui blockchain.

Transform-domain methods simplify analysis and computation, making them central to scientific computing and signal processing. However, existing adaptive strategies often introduce new data structures or require substantial workflow redesign, limiting efficient execution on massively parallel hardware. Here we present spectral dynamic block activation (sp-DBA), an adaptive acceleration framework for transform-domain workflows. sp-DBA dynamically activates transform-domain computation blocks as a calculation evolves, concentrating computation where it is needed while retaining numerical accuracy and optimized transform operations. Across representative workflows in materials science, biology, and optical communications, the demonstrated implementations achieve speedups of up to 28.1-fold for transform-domain updates and up to 8.4-fold overall acceleration; strong- and weak-scaling tests on up to eight GPUs show that the local adaptive update remains effective within distributed FFT workflows. By introducing execution-level adaptivity into mature transform-domain workflows without rebuilding existing solvers, sp-DBA extends the scale, duration, and complexity of scientific simulations and signal-processing calculations on modern parallel systems.

Distributing an AI computation across the GPUs of a multi-GPU server is one of the central problems in systems-for-AI. We present Einsummable, a prototype system that accepts a PyTorch-like description of an AI computation and automatically distributes it across a multi-GPU server, with no device assignments, sharding annotations, or communication operations written by the programmer. Einsummable models every operation as a relational join followed by an aggregation over tensor relations, in which the tuples contain sub-tensors. Each operation exposes its possible decompositions through what we call "join-agg specs". An optimizer then selects decompositions across the whole computation to minimize a communication-cost proxy. Because it searches decompositions rather than a menu of named strategies, Einsummable discovers plans that mesh-based auto-parallelizers cannot. Each decomposed operation is implemented by synthesizing an exchange program, which is a topology-aware generalization of Volcano's exchange operator. Einsummable invokes no canned collectives: all communication and aggregation is special-purpose, derived at compile time. Despite being fully automatic, Einsummable can outperform custom-designed implementations. For example, on LLaMA transformer blocks on an eight-GPU A100 server, Einsummable achieves a geometric-mean runtime of 8.97 ms, versus 13.80 ms for hand-tuned PyTorch and 15.90 ms for vLLM.

This work introduces JuPyLive, a migration mechanism that enables seamless transition of Jupyter notebooks between local resources of user's workstation and remote resources of high-performance computing~(HPC) environments, while preserving the user experience. JuPyLive eliminates the underlying complexities of migration process, enabling users to freely choose among available local and remote resources, directly within the familiar Jupyter notebook environment via a single click. JuPyLive leverages ElasticNotebook to manage in-memory state migration, it automates resource allocation on HPC cluster and orchestrates required remote communication channels between the source and destination to enable a bidirectional migration. Furthermore, HPC status monitor of JuPyLive provides a live overview of available remote resources, allowing users to make informed decisions on choosing the relevant resources before initiating a migration process. The proposed fully automatic mechanism requires no code changes or configurations by the end user, nor does it demand users to learn a new syntax, instead the migration process can be intuitively initiated and monitored using visual elements from within the Jupyter notebook. By bridging the gap between local workspace and remote resources, JuPyLive offers a seamless experience for scaling local resource-intensive workflows with minimal user intervention, thus further democratizing the usage of HPC clusters among the interdisciplinary researchers.

Tensor-Train (TT) decomposition effectively compresses large embedding tables in recommendation models, but TT-based embedding lookup remains inefficient because partially shared computation flows across input indices are not fully reused and intermediate results are repeatedly materialized off-chip between sequential TT-core contractions. We present FlowTT, a flow-aware GPU execution framework that reformulates TT gather as a set of prefix-shared irregular computation flows. FlowTT combines flow-aligned prefix-based index grouping, a fused TT-embedding execution path with on-chip intermediate retention, and persistent-thread scheduling with chunk-based work stealing and L2 checkpointing to preserve reuse under skewed workloads. By co-designing task formation, data buffering, and scheduling with the structure of TT gather, FlowTT reduces redundant TT-core operations, global-memory traffic, and load imbalance. On Meta's synthetic recommendation benchmarks (Meta-240, Meta-480, and Meta-788), FlowTT consistently achieves the lowest latency compared to existing methods. At batch size 32,768, it reduces latency by up to 42.2% in inference and 49.2% in training relative to EcoRec, while also achieving the lowest inference peak memory usage. These results show that exposing prefix-shared computation is key to efficient TT-based embedding execution.

Constant optimization refines the numerical coefficients of candidate expressions in tree-based genetic programming for symbolic regression. But its per-generation cost has led modern GPU-accelerated frameworks to omit it or restrict it to lightweight forms. We present a GPU-resident, batched Levenberg–Marquardt solver that optimizes constants across a structurally heterogeneous population of expression trees using a fixed number of population-wide CUDA launches per iteration. Reverse-mode automatic differentiation assembles the per-tree Jacobian in one backward sweep, making the dominant per-iteration cost independent of the number of constants per tree, and a double-precision delivery guard guarantees that returned constants are never worse than their initial values. On early-generation populations, the solver sustains up to $5.1{\times}10^{5}$ trees per second on an NVIDIA A100; at a GPU-saturated benchmark configuration it delivers roughly $9.9{\times}$ the throughput of Operon running on a 64-core EPYC 7763, while matching fp64-reference quality. Integrated in-process into EvoGP, the solver enables end-to-end search to recover governing equations on $10$ of $18$ constructed problems versus 0 for stock EvoGP. Our code is at https://github.com/TensorConv/CuSR.

Concurrent multi-agent workflows expose future dependencies and serving-state requirements while running on heterogeneous GPU pools with time-varying load, model residency, and resource availability. The logical workflow defines the required computation, whereas its physical scheduling units, model-lifecycle actions, resource ordering, and placement must be selected according to the observed pool state. We present a prediction-guided runtime that uses workflow forecasts to construct and optimize a physical execution graph. Predictor estimates device-specific activation latency, peak memory, and model-loading cost, then propagates these predictions through workflow dependencies to forecast activation readiness and future model demand. Constructor builds semantics-preserving fusion and model-lifecycle alternatives, while Scheduler jointly optimizes their selection, placement, and execution order based on the live pool state. Across a workload spanning three workflow scenarios on a heterogeneous GPU pool, our system reduces end-to-end makespan and overall p95 completion latency under burst arrivals by up to 36.8% and 25.9%, respectively, over state-of-the-art workflow schedulers. It also saves up to 24.63 GPU-s per completed session.

Collaborative inference pools distributed resources to run compute-intensive Vision Transformers (ViTs) in satellite edge computing. Model partitioning enables such collaboration by assigning consecutive layer groups to different nodes, but the large volume of intermediate activation data incurs substantial transfer overhead that can erase its benefit. Token compression reduces downstream computation and activation transfer, but its quality impact depends on input content, model depth, and earlier pruning decisions, while layer offloading must adapt to time-varying contact and battery conditions. We present \sys, a content-aware hierarchical scheduler that screens constellation-wide options to retain a bounded candidate set, then refines each candidate into a complete token compression and layer offloading trajectory using quality prediction and joint planning. A unified objective balances per-task latency, energy, and quality loss against accumulated workload and battery pressures. We implement \sys on an NVIDIA Jetson AGX Orin hardware-in-the-loop testbed and use its validated execution model for constellation-scale trace replay across multiple ViT workloads and constellation settings. At \(5\)~tasks/s, \sys accomplishes 91.6% of released tasks, 26.1 percentage points above MARATD3, the strongest baseline, while reducing mean latency and battery draw by 53.0% and 70.8%, respectively, and meeting quality targets.

Existing concurrency control protocols either introduce nondeterminism, resulting in a serial execution-replay dependency between primary and replica nodes, or rely on impractical prior knowledge of transaction read-write sets. In this paper, we present Lantern, a deterministic concurrency control protocol tailored for high-performance transaction processing systems operating without prior knowledge. The key insight of Lantern is that all zero-out-degree transaction vertices in the local dependency graph can be safely committed in ascending order using an overwrite-permissive strategy. We further introduce a novel Back-Propagation mechanism that iteratively propagates dependency states from sink to source vertices to identify additional committable transactions. We also propose Conflict-Free Batch Selection (CFBS) for read-modify-write intensive scenarios. We integrate Lantern into the open-source blockchain platform ChainMaker. Extensive evaluations on YCSB and SmallBank benchmarks demonstrate that Lantern achieves up to a 4.2x throughput speedup over Aria and improves the throughput of ChainMaker's execution layer by at least 2.2x.

Load shifting across geographic regions to chase intermittent renewable energy availability is commonly used in reducing cloud infrastructure carbon footprint. However, it often omits low-latency applications due to high latency variances of wide area networks (WAN) that interconnect regions. This paper addresses accommodating low-latency applications into load shifting by minimizing their shifting across the WAN. We propose a technique using a hardware-software co-design approach. At the hardware level, we conduct server load matching over renewables supply peaks and valleys by deep idling physical cores in two otherwise identical server pools, with one enabling simultaneous multi-threading (SMT) in CPUs. In return, we achieve a static set of logical cores amidst energy supply dynamics, reducing the probability of workload shifting. At the software level, we efficiently chase the static set of cores for low-latency applications within regions while prioritizing best-effort applications to accommodate shifting requirements across WANs. Our approach exploits the lower performance compromise of SMT cores due to their hardware multi-threading. We implement the proposed technique with OpenStack and CPU idle states and evaluate its performance on a real experimental testbed with Azure VM arrival traces. Results show an 80% reduction in offloading low-latency VMs and a 43.81% reduction in coefficient of variation of p90 end-user latency while having a worst-case latency compromise of 11.97% due to SMT cores.

Proliferation of cloud-based latency-sensitive workloads requires infrastructures tuned to their workload-specific latency constraints. Today, they shape the cloud from a generalized computing platform to diverse workload-specific cloud environments. As the demand for latency-sensitive workloads increases, cloud service providers continue to scale their infrastructure, adversely increasing the carbon footprint and challenging climate-crisis-driven net-zero emission goals. Due to performance-oriented rigid deployment patterns of latency-optimizations, reducing its carbon footprint is challenging. Therefore, efficient techniques that exploit application specific opportunities are needed in that. To this end, we present a detailed taxonomy of recent literature on carbon-aware resource management in latency-sensitive cloud computing environments. Using the taxonomy, we analyze existing works discussing their optimization aspects, identify the gaps, and highlight future research directions.

Humanoid robots are becoming an important part of embodied artificial intelligence, driven by advances in reinforcement learning for locomotion, world models for prediction, and vision-language-action models for general control. However, most of these systems remain static after deployment. A policy is trained offline for a fixed objective and then frozen, even though the tasks, environments, and robot bodies keep drifting over time. An emerging paradigm of self-evolving agents aims to address this problem by allowing systems to improve from their own post-deployment experience. Since most existing studies focus on disembodied software agents, this survey examines how self-evolution changes when an agent has a physical body. We first define self-evolution for humanoids and represent a deployed robot using a state tuple that includes its policy, perception, memory, workflow, and body. This state is updated by an evolution operator in a slow outer loop with a lifelong objective. We then organize the literature into four complementary mechanisms of self-evolution, presented in increasing order of autonomy: self-learning, self-adaptation, self-optimization, and self-generation. Since changes to a humanoid can introduce physical hazards, we treat safety and uncertainty as key design dimensions of the evolution operator, and further formulate admissible evolution as a constraint enforced by a world-model verification gate within a human-oversight envelope. Finally, we present that evaluation should track the robot's evolving trajectory rather than a fixed checkpoint, and we identify the lack of a benchmark designed specifically for self-evolving humanoids. Moreover, we outline open challenges spanning AI algorithms, on-board systems, and governance.

Long-context reasoning for large language models (LLMs) is becoming increasingly important, but training over long sequences remains challenging due to massive memory and communication requirements. Sequence parallelism has emerged as an essential technique for addressing bottlenecks in long sequence LLM training. However, we observe that existing sequence parallelism methods are batch-agnostic and apply uniform sequence partitioning across all batch sizes, resulting in inefficient communication. In this paper, we introduce Batch- Aware Sequence Parallelism (BASP), a sequence parallelism approach that leverages batch structure to reduce communication overhead. BASP exploits batch structure by partitioning GPUs into disjoint sequence-parallel groups according to the micro- batch size. This design reduces the all-to-all communication group size, thereby localizing communication and improving training efficiency. Experimental results on an NVIDIA A100 cluster show that BASP improves end-to-end training time by up to 1.17 - 1.31x in Llama and Qwen models compared to standard sequence parallel baselines, while preserving identical model accuracy and memory usage.

Emerging edge, autonomous, and cyber-physical systems increasingly require mathematical computation across heterogeneous devices connected by unreliable communication networks. Traditional high-performance computing and distributed data-processing frameworks provide powerful abstractions for managed environments but are less suited to decentralized settings where centralized coordination, reliable communication, and global synchronization cannot be assumed. This paper presents Skywing, an open-source platform for decentralized mathematical computing in unreliable environments. Its programming model consists of three abstractions: agents represent participants in a decentralized computation, processors encapsulate algorithm-specific update rules, and iterations manage distributed execution. Skywing supports asynchronous operation, publish-subscribe communication, managed message handling, and the composition of independent algorithms into complex decentralized workflows. We demonstrate Skywing using representative algorithms from consensus, optimization, and numerical linear algebra. Experiments on the native Skywing runtime include Push Sum and Max Consensus, a composed monitoring and control workflow, resilient Push Sum under delayed communication, and resilient asynchronous Jacobi under malevolent data corruption. These demonstrations show that Skywing supports diverse decentralized algorithms while separating mathematical logic from communication and execution infrastructure. Skywing serves as both a deployment framework for decentralized applications and a research platform for developing resilient mathematical algorithms.

Bitcoin's proof-of-work (PoW)-based protocol is remarkable for how little it asks of its participants. Not only can miners take breaks from work whenever they please, but it is almost unique in offering a path of contrition: corrupt miners can reclaim honest status simply by resuming mining on the longest chain. The protocol only requires that honest miners hold the majority of computational power at any given time. Analogous proof-of-stake (PoS) protocols, usually formalized via the sleepy model of Pass and Shi (2017), have fallen short of matching this robustness. In fact, sleepy consensus protocols in the plain PKI model must heavily restrict fluctuations in adversarial participation over time. The recent work of Efron, Neu, and Pitassi (2025) enables fully fluctuating participation in the sleepy model by introducing the external adversary model. Their protocol, however, relies on verifiable delay functions (VDFs), a strong cryptographic primitive that somewhat resembles PoW, by assuming that the adversary cannot compute sequential work significantly faster than honest nodes. In this work, we design a sleepy consensus protocol for fully fluctuating participation with an external adversary under an honest majority, from minimal assumptions: a public key infrastructure (PKI) and a verifiable random function (VRF). In particular, we make no VDF or hardware assumptions. Our key technique is graded wakeness, a novel primitive that allows nodes to form consistent opinions on which other nodes are awake. We further extend our protocol to handle uncorruption, where corrupt nodes return to honesty. This extension requires only a mild additional assumption on the unpredictability of VRF outputs for liveness.

Ordinary differential equations (ODEs) underlie models in science and engineering, and many applications need derivatives of their solutions with respect to parameters. Ensembles of independent trajectories suit graphics processing units (GPUs), but current GPU software forces a trade-off: the fastest ensemble solvers cannot be differentiated in reverse mode at the speed they solve, and the solvers built for differentiation solve more slowly. No single tool has yet offered a reverse-mode gradient at the speed of a fused-kernel solve. We present GRADSOLVE, an open-source JAX library for solving and reverse-mode differentiating low-dimensional ODE ensembles on NVIDIA GPUs. It records the steps an adaptive solver accepts and differentiates a fixed-step replay of them; the returned gradient is the exact discrete adjoint of those steps, the same derivative Diffrax returns by default, obtained more cheaply from a fixed-length chain than from an adaptive loop. It targets ensembles differentiated many times against one recorded mesh, keeps Diffrax as a fallback, and supports explicit and Rosenbrock integrators. Used as a solver, GRADSOLVE's forward-only kernel ran 2.8x faster than DiffEqGPU.jl; used for gradients, once a record exists, it computed them 5.6-14.1x faster than Diffrax's checkpointed adjoint at matched forward-state accuracy across three GPU generations, the advantage narrowing on large ensembles and, on stiff systems, down to parity at tight accuracy. GRADSOLVE is released at https://github.com/ECLIPSE-AI4Science/gradsolve.

Revocation completion, clean state, or operation success can leave authorized work able to cause an effect the application rejects while the provider stays within its contract. We call the absence of all such paths policy-relative effect closure, or effect closure for short. Thus, a grant is closed when its existing authorizations retain no such path, and it cannot issue any new ones. We present EFFECTBOUND, which uses an evidence-supported finite contract to decide whether an interface can truthfully report closure while required work completes. It reduces this to finite control with hidden state and returns a strategy, an impossibility certificate, or no verdict when evidence is insufficient. Machine-checked proofs establish the reduction and checker soundness; the checker derives closure results and validates certificates. Across GitHub, Kubernetes, NATS, and Kafka, closure fails in three ways: an interface lacks a needed control, clean visible state hides active work, or the model stops before the effect frontier—the last point where the effect can be prevented. The GitHub tool cannot bind a merge to the reviewed commit; a controlled run confirms that it may merge a different commit. NATS can report no stored or pending messages while dispatched work can still publish downstream. In Kafka, all fixed-set brokers had applied the revocation, yet an earlier authorized request could still append. We add a gate that blocks new use of revoked authority and delays return until earlier in-flight work completes. In a fixed-set Kafka~4.3.1 test deployment, this closes the studied synchronous, nontransactional write path without blocking unrelated requests. For a grant, authorization ends only when issuance stops and no earlier authorization can reach an effect the application rejects.

Semiconductor quantum-dot arrays are a compelling platform for scalable quantum technologies, yet their practical operation is hindered by the complexity of tuning large-scale devices. Existing automation tools rely on simplified physical models—such as constant-capacitance approximations and equilibrium Hubbard models—which assume instantaneous relaxation to a steady state. These frameworks fail in experimentally critical regimes where measurement rates exceed tunneling dynamics, necessitating more sophisticated non-equilibrium control strategies. To bridge this gap, we introduce QArray+, an extension of the QArray framework that incorporates gate-dependent tunnel coupling and a quantum open-system description of dissipative processes. This approach enables the unified simulation of coherent interdot charge-state hybridization and the non-equilibrium latching dynamics essential for training robust machine-learning models for automated device operation. Implemented in JAX with GPU acceleration, QArray+ scales across GPUs and multi-node systems. For example, a charge stability diagram for a 100X100 grid of gate voltages over 64 dots can be computed in $\sim0.17\,\mathrm{s}$ on multiple GPUs. Since interdot interactions are short-ranged and the corresponding tuning corrections are local, simulations at these scales capture the physics relevant to even larger devices. These capabilities support high-throughput dataset generation for automated device tuning.

Deploying Large Language Models (LLMs) on edge devices typically relies on model compression or split inference. However, compression degrades reasoning capabilities, while split inference suffers from severe Wide Area Network (WAN) communication bottlenecks. Edge-cloud speculative decoding emerges as a promising alternative, leveraging an edge small model to draft tokens for cloud verification. Yet, over volatile WANs, inevitable prediction rejections trigger catastrophic pipeline stalls and network-wide rollbacks, neutralizing collaborative gains. To overcome this, we propose AceSpec, an asymmetric edge-cloud collaborative framework. AceSpec utilizes un-saturated edge compute to proactively construct a probabilistic state cache, effectively transforming network-wide pipeline flushes into $\mathcal{O}(1)$ local memory lookups. To preserve bandwidth, it employs an asymmetric communication protocol that transmits minimal main-chain indices uplink and compact sparse distributions downlink. Furthermore, we introduce a network-aware, Lagrangian-optimized resource allocation strategy that dynamically maximizes the local cache hit rate. Evaluations demonstrate that AceSpec achieves up to a 3.52$\times$ throughput speedup and exhibits exceptional bandwidth immunity, sustaining near-peak inference performance even under severely constrained 50 Kbps WAN conditions.

Semantic triggers in federated learning (FL) can be less conspicuous than synthetic patches, but sample-dependent placement may weaken backdoor implantation across aggregation rounds. This challenge is compounded in decentralized FL (DFL), where topology-dependent peer aggregation repeatedly mixes local models. CACTUS converts label-consistent semantic pairs into target-directed representation shifts. Mask-guided, modality-specific operators isolate trigger effects, couple them across samples, and apply the shifts counterfactually to clean non-target embeddings before peer aggregation. Experiments cover speech, text, tabular, and image tasks under nine aggregation rules. With 30% malicious nodes, CACTUS reaches a nine-rule mean attack success rate (ASR) of 51.2% on Speech Commands and the highest nine-rule mean ASR among evaluated attacks on three of four modalities. Sensitivity analyses show that ASR varies with network topology and increases with the malicious-node ratio. These results indicate that CACTUS can propagate backdoors through repeated DFL aggregation.