This tutorial provides a concise introduction to modern causal modeling by integrating potential outcomes and graphical methods. We motivate causal questions such as counterfactual reasoning under interventions and define binary treatments and potential outcomes. We discuss causal effect measures-including average treatment effects on the treated and on the untreated-and choices of effect scales for binary outcomes. We derive identification in randomized experiments under exchangeability and consistency, and extend to stratification and blocking designs. We present inverse probability weighting with propensity score estimation and robust inference via sandwich estimators. Finally, we introduce causal graphs, d-separation, the backdoor criterion, single-world intervention graphs, and structural equation models, showing how graphical and potential-outcome approaches complement each other. Emphasis is placed on clear notation, intuitive explanations, and practical examples for applied researchers.
Recordings gathered with child-worn devices promised to revolutionize both fundamental and applied speech sciences by allowing the effortless capture of children's naturalistic speech environment and language production. This promise hinges on speech technologies that can transform the sheer mounds of data thus collected into usable information. This paper demonstrates several obstacles blocking progress by summarizing three years' worth of experiments aimed at improving one fundamental task: Voice Type Classification. Our experiments suggest that improvements in representation features, architecture, and parameter search contribute to only marginal gains in performance. More progress is made by focusing on data relevance and quantity, which highlights the importance of collecting data with appropriate permissions to allow sharing.
Integrated Sensing And Communication (ISAC) has been identified as a key 6G application by ITU and 3GPP, with standardization efforts already underway. Sensing tasks, such as target localization, demand more precise characterization of the sensing target (ST) in ISAC channel modeling. The ST couples complexly with environmental scatterers, potentially blocking some multipaths and generating new ones, resulting in power variations compared to the original channel. To accurately model this effect, this paper proposes a coupled ISAC channel model based on measurements and validates it through similarity analysis between simulated and measured channels. In this work, we first conduct ISAC channel measurements in an indoor factory scenario at 105 GHz, where the multipath power variations caused by the ST's interaction with the environment are clearly observed. Then, we propose an ISAC channel modeling framework that incorporates two novel parameters: the Blockage-Region Coupling Factor (BR-CF) and the Forward-Scattering (FS)-CF, which characterize the spatial region and intensity of the coupling effect, respectively. Finally, the proposed model is validated through similarity comparison with measured data, demonstrating higher accuracy for both LoS and NLoS scenarios compared to the non-coupled model. This realistic ISAC channel model provides an effective framework for capturing the ST-environment coupling effect, supporting the design and evaluation of ISAC technologies.
Training large language models is generally done via optimization methods on clusters containing tens of thousands of accelerators, communicating over a high-bandwidth interconnect. Scaling up these clusters is expensive and can become impractical, imposing limits on the size of models that can be trained. Several recent studies have proposed training methods that are less communication intensive, avoiding the need for a highly connected compute cluster. These state-of-the-art low communication training methods still employ a synchronization step for model parameters, which, when performed over all model replicas, can become costly on a low-bandwidth network. In this work, we propose a novel optimization method, NoLoCo, that does not explicitly synchronize all model parameters during training and, as a result, does not require any collective communication. NoLoCo implicitly synchronizes model weights via a novel variant of the Nesterov momentum optimizer by partially averaging model weights with a randomly selected other one. We provide both a theoretical convergence analysis for our proposed optimizer as well as empirical results from language model training. We benchmark NoLoCo on a wide range of accelerator counts and model sizes, between 125M to 6.8B parameters. Our method requires significantly less communication overhead than fully sharded data parallel training or even widely used low communication training method, DiLoCo. The synchronization step itself is estimated to be one magnitude faster than the all-reduce used in DiLoCo for few hundred accelerators training over the internet. We also do not have any global blocking communication that reduces accelerator idling time. Compared to DiLoCo, we also observe up to $4\%$ faster convergence rate with wide range of model sizes and accelerator counts.
Understanding the plausible upper bounds of extreme weather events is essential for risk assessment in a warming climate. Existing methods, based on large ensembles of physics-based models, are often computationally expensive or lack the fidelity needed to simulate rare, high-impact extremes. Here, we present a novel framework that leverages a differentiable hybrid climate model, NeuralGCM, to optimize initial conditions and generate physically consistent worst-case heatwave trajectories. Applied to the 2021 Pacific Northwest heatwave, our method produces temperature anomalies up to 3.7 $^\circ$C above the most extreme member of a 75-member ensemble. These trajectories feature intensified atmospheric blocking and amplified Rossby wave patterns--hallmarks of severe heat events. Our results demonstrate that differentiable climate models can efficiently explore the upper tails of event likelihoods, providing a powerful new approach for constructing targeted storylines of extreme weather under climate change.
The rise of Large Language Models (LLMs) has driven progress in reasoning tasks -- from program synthesis to scientific hypothesis generation -- yet their ability to handle ranked preferences and structured algorithms in combinatorial domains remains underexplored. We study matching markets, a core framework behind applications like resource allocation and ride-sharing, which require reconciling individual ranked preferences to ensure stable outcomes. We evaluate several state-of-the-art models on a hierarchy of preference-based reasoning tasks -- ranging from stable-matching generation to instability detection, instability resolution, and fine-grained preference queries -- to systematically expose their logical and algorithmic limitations in handling ranked inputs. Surprisingly, even top-performing models with advanced reasoning struggle to resolve instability in large markets, often failing to identify blocking pairs or execute algorithms iteratively. We further show that parameter-efficient fine-tuning (LoRA) significantly improves performance in small markets, but fails to bring about a similar improvement on large instances, suggesting the need for more sophisticated strategies to improve LLMs' reasoning with larger-context inputs.
We introduce a method to reduce refusal rates of large language models (LLMs) on sensitive content without modifying model weights or prompts. Motivated by the observation that refusals in certain models were often preceded by the specific token sequence of a token marking the beginning of the chain-of-thought (CoT) block (<think>) followed by a double newline token (\n\n), we investigate the impact of two simple formatting adjustments during generation: suppressing \n\n after <think> and suppressing the end-of-sequence token after the end of the CoT block (</think>). Our method requires no datasets, parameter changes, or training, relying solely on modifying token probabilities during generation. In our experiments with official DeepSeek-R1 distillations, these interventions increased the proportion of substantive answers to sensitive prompts without affecting performance on standard benchmarks. Our findings suggest that refusal behaviors can be circumvented by blocking refusal subspaces at specific points in the generation process.
The interpretability of Mixture-of-Experts (MoE) models, especially those with heterogeneous designs, remains underexplored. Existing attribution methods for dense models fail to capture dynamic routing-expert interactions in sparse MoE architectures. To address this issue, we propose a cross-level attribution algorithm to analyze sparse MoE architectures (Qwen 1.5-MoE, OLMoE, Mixtral-8x7B) against dense models (Qwen 1.5-7B, Llama-7B, Mixtral-7B). Results show MoE models achieve 37% higher per-layer efficiency via a "mid-activation, late-amplification" pattern: early layers screen experts, while late layers refine knowledge collaboratively. Ablation studies reveal a "basic-refinement" framework--shared experts handle general tasks (entity recognition), while routed experts specialize in domain-specific processing (geographic attributes). Semantic-driven routing is evidenced by strong correlations between attention heads and experts (r=0.68), enabling task-aware coordination. Notably, architectural depth dictates robustness: deep Qwen 1.5-MoE mitigates expert failures (e.g., 43% MRR drop in geographic tasks when blocking top-10 experts) through shared expert redundancy, whereas shallow OLMoE suffers severe degradation (76% drop). Task sensitivity further guides design: core-sensitive tasks (geography) require concentrated expertise, while distributed-tolerant tasks (object attributes) leverage broader participation. These insights advance MoE interpretability, offering principles to balance efficiency, specialization, and robustness.
CAPTCHAs have been a critical bottleneck for deploying web agents in real-world applications, often blocking them from completing end-to-end automation tasks. While modern multimodal LLM agents have demonstrated impressive performance in static perception tasks, their ability to handle interactive, multi-step reasoning challenges like CAPTCHAs is largely untested. To address this gap, we introduce Open CaptchaWorld, the first web-based benchmark and platform specifically designed to evaluate the visual reasoning and interaction capabilities of MLLM-powered agents through diverse and dynamic CAPTCHA puzzles. Our benchmark spans 20 modern CAPTCHA types, totaling 225 CAPTCHAs, annotated with a new metric we propose: CAPTCHA Reasoning Depth, which quantifies the number of cognitive and motor steps required to solve each puzzle. Experimental results show that humans consistently achieve near-perfect scores, state-of-the-art MLLM agents struggle significantly, with success rates at most 40.0% by Browser-Use Openai-o3, far below human-level performance, 93.3%. This highlights Open CaptchaWorld as a vital benchmark for diagnosing the limits of current multimodal agents and guiding the development of more robust multimodal reasoning systems. Code and Data are available at this https URL.
Geomagnetic disturbances are a threat to the reliability and security of our national critical energy infrastructures. These events specifically result in geomagnetically induced currents, which can cause damage to transformers due to magnetic saturation. In order to mitigate these effects, blocker devices must be placed in optimal locations. Finding this placement requires a dc representation of the ac transmission lines, which this paper discusses. Different decisions in this process, including the method of representing the blocking devices, result in significant variations to the power loss calculations. To analyze these effects, we conclude the paper by comparing the losses on a sample network with different modeling implementations.
In cloud-native systems, Kubernetes clusters with interdependent services often face challenges to their operational resilience due to poor workload management issues such as resource blocking, bottlenecks, or continuous pod crashes. These vulnerabilities are further amplified in adversarial scenarios, such as Distributed Denial-of-Service attacks (DDoS). Conventional Horizontal Pod Autoscaling (HPA) approaches struggle to address such dynamic conditions, while reinforcement learning-based methods, though more adaptable, typically optimize single goals like latency or resource usage, neglecting broader failure scenarios. We propose decomposing the overarching goal of maintaining operational resilience into failure-specific sub-goals delegated to collaborative agents, collectively forming an HPA Multi-Agent System (MAS). We introduce an automated, four-phase online framework for HPA MAS design: 1) modeling a digital twin built from cluster traces; 2) training agents in simulation using roles and missions tailored to failure contexts; 3) analyzing agent behaviors for explainability; and 4) transferring learned policies to the real cluster. Experimental results demonstrate that the generated HPA MASs outperform three state-of-the-art HPA systems in sustaining operational resilience under various adversarial conditions in a proposed complex cluster.
Training multi-modal large language models (MLLMs) that align with human intentions is a long-term challenge. Traditional score-only reward models for alignment suffer from low accuracy, weak generalization, and poor interpretability, blocking the progress of alignment methods, e.g., reinforcement learning from human feedback (RLHF). Generative reward models (GRMs) leverage MLLMs' intrinsic reasoning capabilities to discriminate pair-wise responses, but their pair-wise paradigm makes it hard to generalize to learnable rewards. We introduce Generative RLHF-V, a novel alignment framework that integrates GRMs with multi-modal RLHF. We propose a two-stage pipeline: $\textbf{multi-modal generative reward modeling from RL}$, where RL guides GRMs to actively capture human intention, then predict the correct pair-wise scores; and $\textbf{RL optimization from grouped comparison}$, which enhances multi-modal RL scoring precision by grouped responses comparison. Experimental results demonstrate that, besides out-of-distribution generalization of RM discrimination, our framework improves 4 MLLMs' performance across 7 benchmarks by $18.1\%$, while the baseline RLHF is only $5.3\%$. We further validate that Generative RLHF-V achieves a near-linear improvement with an increasing number of candidate responses. Our code and models can be found at https://generative-rlhf-v.github.io.
In this paper, channel estimation (CE) for uplink hybrid-field communications involving multiple Internet of Things (IoT) devices assisted by an active intelligent reflecting surface (IRS) is investigated. Firstly, to reduce the complexity of near-field (NF) channel modeling and estimation between IoT devices and active IRS, a sub-blocking strategy for active IRS is proposed. Specifically, the entire active IRS is divided into multiple smaller sub-blocks, so that IoT devices are located in the far-field (FF) region of each sub block, while also being located in the NF region of the entire active IRS. This strategy significantly simplifies the channel model and reduces the parameter estimation dimension by decoupling the high-dimensional NF channel parameter space into low dimensional FF sub channels. Subsequently, the relationship between channel approximation error and CE error with respect to the number of sub blocks is derived, and the optimal number of sub blocks is solved based on the criterion of minimizing the total error. In addition, considering that the amplification capability of active IRS requires power consumption, a closed-form expression for the optimal power allocation factor is derived. To further reduce the pilot overhead, a lightweight CE algorithm based on convolutional autoencoder (CAE) and multi-head attention mechanism, called CAEformer, is designed. The Cramer-Rao lower bound is derived to evaluate the proposed algorithm's performance. Finally, simulation results demonstrate the proposed CAEformer network significantly outperforms the conventional least square and minimum mean square error scheme in terms of estimation accuracy.
We propose ELIS, a serving system for Large Language Models (LLMs) featuring an Iterative Shortest Remaining Time First (ISRTF) scheduler designed to efficiently manage inference tasks with the shortest remaining tokens. Current LLM serving systems often employ a first-come-first-served scheduling strategy, which can lead to the "head-of-line blocking" problem. To overcome this limitation, it is necessary to predict LLM inference times and apply a shortest job first scheduling strategy. However, due to the auto-regressive nature of LLMs, predicting the inference latency is challenging. ELIS addresses this challenge by training a response length predictor for LLMs using the BGE model, an encoder-based state-of-the-art model. Additionally, we have devised the ISRTF scheduling strategy, an optimization of shortest remaining time first tailored to existing LLM iteration batching. To evaluate our work in an industrial setting, we simulate streams of requests based on our study of real-world user LLM serving trace records. Furthermore, we implemented ELIS as a cloud-native scheduler system on Kubernetes to evaluate its performance in production environments. Our experimental results demonstrate that ISRTF reduces the average job completion time by up to 19.6%.
This paper presents a mathematical framework for characterizing state blocking in discrete event systems (DES) under transition deletions. We introduce a path-based analysis approach that determines whether systems maintain non-blocking properties when transitions are removed. Through formal analysis and case studies, we establish three key contributions: a mathematical characterization of transition-induced blocking with necessary and sufficient conditions, a definition of robust deviations that preserve non-blocking properties, and an algorithm for identifying critical transitions and analyzing system behavior under deletions. Our algorithm reduces computational complexity by leveraging minimal blocking sets, achieving significant reduction in computational requirements. We demonstrate the framework's effectiveness through manufacturing system and autonomous vehicle case studies, showing substantial improvements in identifying critical transitions and predicting potential blocking scenarios across different application domains.
Many drugs have been withdrawn from the market worldwide, at a cost of billions of dollars, because of patient fatalities due to them unexpectedly disturbing heart rhythm. Even drugs for ailments as mild as hay fever have been withdrawn due to an unacceptable increase in risk of these heart rhythm disturbances. Consequently, the whole pharmaceutical industry expends a huge effort in checking all new drugs for any unwanted side effects on the heart. The predominant root cause has been identified as drug molecules blocking ionic current flows in the heart. Block of individual types of ionic currents can now be measured experimentally at an early stage of drug development, and this is the standard screening approach for a number of ion currents in many large pharmaceutical companies. However, clinical risk is a complex function of the degree of block of many different types of cardiac ion currents, and this is difficult to understand by looking at results of these screens independently. By using ordinary differential equation models for the electrical activity of heart cells (electrophysiology models) we can integrate information from different types of currents, to predict the effect on whole heart cells and subsequent risk of side effects. The resulting simulations can provide a more accurate summary of the risk of a drug earlier in development and hence more cheaply than the pre-existing approaches.
Reconfigurable intelligent surface (RIS) is emerging as a promising technology for next-generation wireless communication networks, offering a variety of merits such as the ability to tailor the communication environment. Moreover, deploying multiple RISs helps mitigate severe signal blocking between the base station (BS) and users, providing a practical and efficient solution to enhance the service coverage. However, fully reaping the potential of a multi-RIS aided communication system requires solving a non-convex optimization problem. This challenge motivates the adoption of learning-based methods for determining the optimal policy. In this paper, we introduce a novel heterogeneous graph neural network (GNN) to effectively leverage the graph topology of a wireless communication environment. Specifically, we design an association scheme that selects a suitable RIS for each user. Then, we maximize the weighted sum rate (WSR) of all the users by iteratively optimizing the RIS association scheme, and beamforming designs until the considered heterogeneous GNN converges. Based on the proposed approach, each user is associated with the best RIS, which is shown to significantly improve the system capacity in multi-RIS multi-user millimeter wave (mmWave) communications. Specifically, simulation results demonstrate that the proposed heterogeneous GNN closely approaches the performance of the high-complexity alternating optimization (AO) algorithm in the considered multi-RIS aided communication system, and it outperforms other benchmark schemes. Moreover, the performance improvement achieved through the RIS association scheme is shown to be of the order of 30%.
The surging demand for GPUs in datacenters for machine learning (ML) has made efficient GPU utilization crucial. However, meeting the diverse needs of ML models while optimizing resource usage is challenging. To enable transparent, fine-grained GPU management that maximizes utilization and energy efficiency while maintaining strong isolation, an operating system (OS) approach is needed. This paper introduces LithOS, a first step toward a GPU OS. LithOS includes the following new abstractions and mechanisms for efficient GPU resource management: (i) a novel TPC Scheduler that supports spatial scheduling at the granularity of individual TPCs, unlocking efficient TPC stealing between workloads; (ii) transparent kernel atomization to reduce head-of-line blocking and enable dynamic resource reallocation mid-execution; (iii) a lightweight hardware right-sizing mechanism that determines the minimal TPC resources needed per atom; and (iv) a transparent power management mechanism that reduces power consumption based on in-flight work behavior. We implement LithOS in Rust and evaluate its performance across extensive ML environments, comparing it to state-of-the-art solutions from NVIDIA and prior research. For inference stacking, LithOS reduces tail latencies by 13x compared to MPS; compared to the best SotA, it reduces tail latencies by 3x while improving aggregate throughput by 1.6x. In hybrid inference-training stacking, LithOS reduces tail latencies by 4.7x compared to MPS; compared to the best SotA, it reduces tail latencies 1.18x while improving aggregate throughput by 1.35x. Finally, for a modest performance hit under 4%, LithOS's right-sizing provides a quarter of GPU capacity savings on average, while for a 7% hit, its power management yields a quarter of a GPU's energy savings. Overall, LithOS increases GPU efficiency, establishing a foundation for future OS research on GPUs.
Objective: Healthcare data fragmentation presents a major challenge for linking patient data, necessitating robust record linkage to integrate patient records from diverse sources. This study investigates the feasibility of leveraging language models for automated patient record linkage, focusing on two key tasks: blocking and matching. Materials and Methods: We utilized real-world healthcare data from the Missouri Cancer Registry and Research Center, linking patient records from two independent sources using probabilistic linkage as a baseline. A transformer-based model, RoBERTa, was fine-tuned for blocking using sentence embeddings. For matching, several language models were experimented under fine-tuned and zero-shot settings, assessing their performance against ground truth labels. Results: The fine-tuned blocking model achieved a 92% reduction in the number of candidate pairs while maintaining near-perfect recall. In the matching task, fine-tuned Mistral-7B achieved the best performance with only 6 incorrect predictions. Among zero-shot models, Mistral-Small-24B performed best, with a total of 55 incorrect predictions. Discussion: Fine-tuned language models achieved strong performance in patient record blocking and matching with minimal errors. However, they remain less accurate and efficient than a hybrid rule-based and probabilistic approach for blocking. Additionally, reasoning models like DeepSeek-R1 are impractical for large-scale record linkage due to high computational costs. Conclusion: This study highlights the potential of language models for automating patient record linkage, offering improved efficiency by eliminating the manual efforts required to perform patient record linkage. Overall, language models offer a scalable solution that can enhance data integration, reduce manual effort, and support disease surveillance and research.
The Deferred Acceptance algorithm (DA) frequently produces Pareto inefficient allocations in school choice problems. While a number of efficient mechanisms that Pareto-dominate DA are available, a normative question remains unexplored: which students should benefit from efficiency enhancements? We address it by introducing the concept of \emph{maximally improvable students}, who benefit in every improvement over DA that includes as many students as possible in set-inclusion terms. We prove that common mechanisms such as Efficiency-Adjusted DA (EADA) and Top Trading Cycles applied to DA (DA+TTC) can fall significantly short of this benchmark. These mechanisms may only improve two maximally-improvable students when up to $n-1$ could benefit. Addressing this limitation, we develop the Maximum Improvement over DA mechanism (MIDA), which generates an efficient allocation that maximises the number of students improved over DA. We show that MIDA can generate fewer blocking pairs than EADA and DA+TTC, demonstrating that its distributional improvements need not come at the cost of high justified envy.