Ravishankar K. Iyer

Saurabh Jha, Shengkun Cui, Zbigniew Kalbarczyk et al.

Despite the successful demonstration of autonomous vehicles (AVs), such as self-driving cars, ensuring AV safety remains a challenging task. Although some actors influence an AV's driving decisions more than others, current approaches pay equal attention to each actor on the road. An actor's influence on the AV's decision can be characterized in terms of its ability to decrease the number of safe navigational choices for the AV. In this work, we propose a safety threat indicator (STI) using counterfactual reasoning to estimate the importance of each actor on the road with respect to its influence on the AV's safety. We use this indicator to (i) characterize the existing real-world datasets to identify rare hazardous scenarios as well as the poor performance of existing controllers in such scenarios; and (ii) design an RL based safety mitigation controller to proactively mitigate the safety hazards those actors pose to the AV. Our approach reduces the accident rate for the state-of-the-art AV agent(s) in rare hazardous scenarios by more than 70%.

Chang Hu, Krishnakant V. Saboo, Ahmad H. Ali et al.

Primary sclerosing cholangitis (PSC) is a rare disease wherein altered bile acid metabolism contributes to sustained liver injury. This paper introduces REMEDI, a framework that captures bile acid dynamics and the body's adaptive response during PSC progression that can assist in exploring treatments. REMEDI merges a differential equation (DE)-based mechanistic model that describes bile acid metabolism with reinforcement learning (RL) to emulate the body's adaptations to PSC continuously. An objective of adaptation is to maintain homeostasis by regulating enzymes involved in bile acid metabolism. These enzymes correspond to the parameters of the DEs. REMEDI leverages RL to approximate adaptations in PSC, treating homeostasis as a reward signal and the adjustment of the DE parameters as the corresponding actions. On real-world data, REMEDI generated bile acid dynamics and parameter adjustments consistent with published findings. Also, our results support discussions in the literature that early administration of drugs that suppress bile acid synthesis may be effective in PSC treatment.

Haoran Qiu, Weichao Mao, Archit Patke et al.

Large language models (LLMs) have been driving a new wave of interactive AI applications across numerous domains. However, efficiently serving LLM inference requests is challenging due to their unpredictable execution times originating from the autoregressive nature of generative models. Existing LLM serving systems exploit first-come-first-serve (FCFS) scheduling, suffering from head-of-line blocking issues. To address the non-deterministic nature of LLMs and enable efficient interactive LLM serving, we present a speculative shortest-job-first (SSJF) scheduler that uses a light proxy model to predict LLM output sequence lengths. Our open-source SSJF implementation does not require changes to memory management or batching strategies. Evaluations on real-world datasets and production workload traces show that SSJF reduces average job completion times by 30.5-39.6% and increases throughput by 2.2-3.6x compared to FCFS schedulers, across no batching, dynamic batching, and continuous batching settings.

Anirudh Choudhary, Mosbah Aouad, Krishnakant Saboo et al.

Squamous cell carcinoma (SCC) is the most common cancer subtype, with an increasing incidence and a significant impact on cancer-related mortality. SCC grading using whole slide images is inherently challenging due to the lack of a reliable protocol and substantial tissue heterogeneity. We propose RACR-MIL, the first weakly-supervised SCC grading approach achieving robust generalization across multiple anatomies (skin, head and neck, lung). RACR-MIL is an attention-based multiple-instance learning framework that enhances grade-relevant contextual representation learning and addresses tumor heterogeneity through two key innovations: (1) a hybrid WSI graph that captures both local tissue context and non-local phenotypical dependencies between tumor regions, and (2) a rank-ordering constraint in the attention mechanism that consistently prioritizes higher-grade tumor regions, aligning with pathologists diagnostic process. Our model achieves state-of-the-art performance across multiple SCC datasets, achieving 3-9% higher grading accuracy, resilience to class imbalance, and up to 16% improved tumor localization. In a pilot study, pathologists reported that RACR-MIL improved grading efficiency in 60% of cases, underscoring its potential as a clinically viable cancer diagnosis and grading assistant.

Shengkun Cui, Rahul Krishna, Saurabh Jha et al.

Cloud incidents pose major operational challenges in production, with unresolved production cloud incidents cost on average over $2M per hour. Prior research identifies code- and configuration-related issues as the predominant category of root causes in cloud incidents. This paper introduces PRAXIS, an orchestrator that manages and deploys an agentic workflow for diagnosing code- and configuration-caused cloud incidents. PRAXIS employs an LLM-driven structured traversal over two types of graph: (1) a service dependency graph (SDG) that captures microservice-level dependencies; and (2) a hammock-block program dependence graph (PDG) that captures code-level dependencies for each microservice. Together, these graphs encode microservice- and code-level dependencies and the LLM acts as a traversal policy over these graphs, moving between services and code dependencies to localize and explain failures. Compared to state-of-the-art ReAct baselines, PRAXIS improves RCA accuracy by up to 3.1x while reducing token consumption by 3.8x. PRAXIS is demonstrated on a set of 30 comprehensive real-world incidents that is being compiled into an RCA benchmark.

Mosbah Aouad, Anirudh Choudhary, Awais Farooq et al.

Pancreatic ductal adenocarcinoma (PDAC) is one of the deadliest cancers, and early detection remains a major clinical challenge due to the absence of specific symptoms and reliable biomarkers. In this work, we propose a new multimodal approach that integrates longitudinal diagnosis code histories and routinely collected laboratory measurements from electronic health records to detect PDAC up to one year prior to clinical diagnosis. Our method combines neural controlled differential equations to model irregular lab time series, pretrained language models and recurrent networks to learn diagnosis code trajectory representations, and cross-attention mechanisms to capture interactions between the two modalities. We develop and evaluate our approach on a real-world dataset of nearly 4,700 patients and achieve significant improvements in AUC ranging from 6.5% to 15.5% over state-of-the-art methods. Furthermore, our model identifies diagnosis codes and laboratory panels associated with elevated PDAC risk, including both established and new biomarkers. Our code is available at https://github.com/MosbahAouad/EarlyPDAC-MML.

Archit Patke, Dhemath Reddy, Saurabh Jha et al.

Large language model (LLM) serving is becoming an increasingly critical workload for cloud providers. Existing LLM serving systems focus on interactive requests, such as chatbots and coding assistants, with tight latency SLO requirements. However, when such systems execute batch requests that have relaxed SLOs along with interactive requests, it leads to poor multiplexing and inefficient resource utilization. To address these challenges, we propose QLM, a queue management system for LLM serving. QLM maintains batch and interactive requests across different models and SLOs in a request queue. Optimal ordering of the request queue is critical to maintain SLOs while ensuring high resource utilization. To generate this optimal ordering, QLM uses a Request Waiting Time (RWT) Estimator that estimates the waiting times for requests in the request queue. These estimates are used by a global scheduler to orchestrate LLM Serving Operations (LSOs) such as request pulling, request eviction, load balancing, and model swapping. Evaluation on heterogeneous GPU devices and models with real-world LLM serving dataset shows that QLM improves SLO attainment by 40-90% and throughput by 20-400% while maintaining or improving device utilization compared to other state-of-the-art LLM serving systems. QLM's evaluation is based on the production requirements of a cloud provider. QLM is publicly available at https://www.github.com/QLM-project/QLM.

Saurabh Jha, Timothy Tsai, Siva Hari et al.

Fully autonomous vehicles (AVs), i.e., AVs with autonomy level 5, are expected to dominate road transportation in the near-future and contribute trillions of dollars to the global economy. The general public, government organizations, and manufacturers all have significant concern regarding resiliency and safety standards of the autonomous driving system (ADS) of AVs . In this work, we proposed and developed (a) `Kayotee' - a fault injection-based tool to systematically inject faults into software and hardware components of the ADS to assess the safety and reliability of AVs to faults and errors, and (b) an ontology model to characterize errors and safety violations impacting reliability and safety of AVs. Kayotee is capable of characterizing fault propagation and resiliency at different levels - (a) hardware, (b) software, (c) vehicle dynamics, and (d) traffic resilience. We used Kayotee to study a proprietary ADS technology built by Nvidia corporation and are currently applying Kayotee to other open-source ADS systems.

Homa Alemzadeh, Daniel Chen, Andrew Lewis et al.

Robotic telesurgical systems are one of the most complex medical cyber-physical systems on the market, and have been used in over 1.75 million procedures during the last decade. Despite significant improvements in design of robotic surgical systems through the years, there have been ongoing occurrences of safety incidents during procedures that negatively impact patients. This paper presents an approach for systems-theoretic safety assessment of robotic telesurgical systems using software-implemented fault-injection. We used a systemstheoretic hazard analysis technique (STPA) to identify the potential safety hazard scenarios and their contributing causes in RAVEN II robot, an open-source robotic surgical platform. We integrated the robot control software with a softwareimplemented fault-injection engine which measures the resilience of the system to the identified safety hazard scenarios by automatically inserting faults into different parts of the robot control software. Representative hazard scenarios from real robotic surgery incidents reported to the U.S. Food and Drug Administration (FDA) MAUDE database were used to demonstrate the feasibility of the proposed approach for safety-based design of robotic telesurgical systems.

Archit Patke, Dhemath Reddy, Saurabh Jha et al.

Large language model (LLM) serving is becoming an increasingly important workload for cloud providers. Based on performance SLO requirements, LLM inference requests can be divided into (a) interactive requests that have tight SLOs in the order of seconds, and (b) batch requests that have relaxed SLO in the order of minutes to hours. These SLOs can degrade based on the arrival rates, multiplexing, and configuration parameters, thus necessitating the use of resource autoscaling on serving instances and their batch sizes. However, previous autoscalers for LLM serving do not consider request SLOs leading to unnecessary scaling and resource under-utilization. To address these limitations, we introduce Chiron, an autoscaler that uses the idea of hierarchical backpressure estimated using queue size, utilization, and SLOs. Our experiments show that Chiron achieves up to 90% higher SLO attainment and improves GPU efficiency by up to 70% compared to existing solutions.

Shengkun Cui, Archit Patke, Hung Nguyen et al.

This study characterizes GPU resilience in Delta HPC, a large-scale AI system that consists of 1,056 A100 and H100 GPUs, with over 1,300 petaflops of peak throughput. Delta HPC is operated by the National Center for Supercomputing Applications (NCSA) at the University of Illinois Urbana-Champaign. We used 2.5 years of operational data (11.7 million GPU hours) on GPU errors. Our major findings include: (i) H100 GPU memory resilience is worse than A100 GPU memory, with 3.2x lower per-GPU MTBE for memory errors, (ii) The GPU memory error-recovery mechanisms on H100 GPUs are insufficient to handle the increased memory capacity, (iii) H100 GPUs demonstrate significantly improved GPU hardware resilience over A100 GPUs with respect to critical hardware components, (iv) GPU errors on both A100 and H100 GPUs frequently result in job failures due to the lack of robust recovery mechanisms at the application level, and (v) We project the impact of GPU node availability on larger-scales and find that significant overprovisioning of 5% is necessary to handle GPU failures.

Saurabh Jha, Yan Miao, Zbigniew Kalbarczyk et al.

Driving in a dynamic environment that consists of other actors is inherently a risky task as each actor influences the driving decision and may significantly limit the number of choices in terms of navigation and safety plan. The risk encountered by the Ego actor depends on the driving scenario and the uncertainty associated with predicting the future trajectories of the other actors in the driving scenario. However, not all objects pose a similar risk. Depending on the object's type, trajectory, position, and the associated uncertainty with these quantities; some objects pose a much higher risk than others. The higher the risk associated with an actor, the more attention must be directed towards that actor in terms of resources and safety planning. In this paper, we propose a novel risk metric to calculate the importance of each actor in the world and demonstrate its usefulness through a case study.

Krishnakant V. Saboo, Anirudh Choudhary, Yurui Cao et al.

We model Alzheimer's disease (AD) progression by combining differential equations (DEs) and reinforcement learning (RL) with domain knowledge. DEs provide relationships between some, but not all, factors relevant to AD. We assume that the missing relationships must satisfy general criteria about the working of the brain, for e.g., maximizing cognition while minimizing the cost of supporting cognition. This allows us to extract the missing relationships by using RL to optimize an objective (reward) function that captures the above criteria. We use our model consisting of DEs (as a simulator) and the trained RL agent to predict individualized 10-year AD progression using baseline (year 0) features on synthetic and real data. The model was comparable or better at predicting 10-year cognition trajectories than state-of-the-art learning-based models. Our interpretable model demonstrated, and provided insights into, "recovery/compensatory" processes that mitigate the effect of AD, even though those processes were not explicitly encoded in the model. Our framework combines DEs with RL for modelling AD progression and has broad applicability for understanding other neurological disorders.

Subho S. Banerjee, Saurabh Jha, Zbigniew T. Kalbarczyk et al.

Hardware performance counters (HPCs) that measure low-level architectural and microarchitectural events provide dynamic contextual information about the state of the system. However, HPC measurements are error-prone due to non determinism (e.g., undercounting due to event multiplexing, or OS interrupt-handling behaviors). In this paper, we present BayesPerf, a system for quantifying uncertainty in HPC measurements by using a domain-driven Bayesian model that captures microarchitectural relationships between HPCs to jointly infer their values as probability distributions. We provide the design and implementation of an accelerator that allows for low-latency and low-power inference of the BayesPerf model for x86 and ppc64 CPUs. BayesPerf reduces the average error in HPC measurements from 40.1% to 7.6% when events are being multiplexed. The value of BayesPerf in real-time decision-making is illustrated with a simple example of scheduling of PCIe transfers.

Subho S Banerjee, Saurabh Jha, Zbigniew T. Kalbarczyk et al.

The problem of scheduling of workloads onto heterogeneous processors (e.g., CPUs, GPUs, FPGAs) is of fundamental importance in modern data centers. Current system schedulers rely on application/system-specific heuristics that have to be built on a case-by-case basis. Recent work has demonstrated ML techniques for automating the heuristic search by using black-box approaches which require significant training data and time, which make them challenging to use in practice. This paper presents Symphony, a scheduling framework that addresses the challenge in two ways: (i) a domain-driven Bayesian reinforcement learning (RL) model for scheduling, which inherently models the resource dependencies identified from the system architecture; and (ii) a sampling-based technique to compute the gradients of a Bayesian model without performing full probabilistic inference. Together, these techniques reduce both the amount of training data and the time required to produce scheduling policies that significantly outperform black-box approaches by up to 2.2x.

Saurabh Jha, Shengkun Cui, Tianyin Xu et al.

We present Kaleidoscope an innovative system that supports live forensics for application performance problems caused by either individual component failures or resource contention issues in large-scale distributed storage systems. The design of Kaleidoscope is driven by our study of I/O failures observed in a peta-scale storage system anonymized as PetaStore. Kaleidoscope is built on three key features: 1) using temporal and spatial differential observability for end-to-end performance monitoring of I/O requests, 2) modeling the health of storage components as a stochastic process using domain-guided functions that accounts for path redundancy and uncertainty in measurements, and, 3) observing differences in reliability and performance metrics between similar types of healthy and unhealthy components to attribute the most likely root causes. We deployed Kaleidoscope on PetaStore and our evaluation shows that Kaleidoscope can run live forensics at 5-minute intervals and pinpoint the root causes of 95.8% of real-world performance issues, with negligible monitoring overhead.

Saurabh Jha, Subho S. Banerjee, Timothy Tsai et al.

The safety and resilience of fully autonomous vehicles (AVs) are of significant concern, as exemplified by several headline-making accidents. While AV development today involves verification, validation, and testing, end-to-end assessment of AV systems under accidental faults in realistic driving scenarios has been largely unexplored. This paper presents DriveFI, a machine learning-based fault injection engine, which can mine situations and faults that maximally impact AV safety, as demonstrated on two industry-grade AV technology stacks (from NVIDIA and Baidu). For example, DriveFI found 561 safety-critical faults in less than 4 hours. In comparison, random injection experiments executed over several weeks could not find any safety-critical faults

Yogatheesan Varatharajah, Brent Berry, Jan Cimbalnik et al.

An ability to map seizure-generating brain tissue, i.e., the seizure onset zone (SOZ), without recording actual seizures could reduce the duration of invasive EEG monitoring for patients with drug-resistant epilepsy. A widely-adopted practice in the literature is to compare the incidence (events/time) of putative pathological electrophysiological biomarkers associated with epileptic brain tissue with the SOZ determined from spontaneous seizures recorded with intracranial EEG, primarily using a single biomarker. Clinical translation of the previous efforts suffers from their inability to generalize across multiple patients because of (a) the inter-patient variability and (b) the temporal variability in the epileptogenic activity. Here, we report an artificial intelligence-based approach for combining multiple interictal electrophysiological biomarkers and their temporal characteristics as a way of accounting for the above barriers and show that it can reliably identify seizure onset zones in a study cohort of 82 patients who underwent evaluation for drug-resistant epilepsy. Our investigation provides evidence that utilizing the complementary information provided by multiple electrophysiological biomarkers and their temporal characteristics can significantly improve the localization potential compared to previously published single-biomarker incidence-based approaches, resulting in an average area under ROC curve (AUC) value of 0.73 in a cohort of 82 patients. Our results also suggest that recording durations between ninety minutes and two hours are sufficient to localize SOZs with accuracies that may prove clinically relevant. The successful validation of our approach on a large cohort of 82 patients warrants future investigation on the feasibility of utilizing intra-operative EEG monitoring and artificial intelligence to localize epileptogenic brain tissue.

Yogatheesan Varatharajah, Brent Berry, Sanmi Koyejo et al.

Clinical trials involving multiple treatments utilize randomization of the treatment assignments to enable the evaluation of treatment efficacies in an unbiased manner. Such evaluation is performed in post hoc studies that usually use supervised-learning methods that rely on large amounts of data collected in a randomized fashion. That approach often proves to be suboptimal in that some participants may suffer and even die as a result of having not received the most appropriate treatments during the trial. Reinforcement-learning methods improve the situation by making it possible to learn the treatment efficacies dynamically during the course of the trial, and to adapt treatment assignments accordingly. Recent efforts using \textit{multi-arm bandits}, a type of reinforcement-learning methods, have focused on maximizing clinical outcomes for a population that was assumed to be homogeneous. However, those approaches have failed to account for the variability among participants that is becoming increasingly evident as a result of recent clinical-trial-based studies. We present a contextual-bandit-based online treatment optimization algorithm that, in choosing treatments for new participants in the study, takes into account not only the maximization of the clinical outcomes but also the patient characteristics. We evaluated our algorithm using a real clinical trial dataset from the International Stroke Trial. The results of our retrospective analysis indicate that the proposed approach performs significantly better than either a random assignment of treatments (the current gold standard) or a multi-arm-bandit-based approach, providing substantial gains in the percentage of participants who are assigned the most suitable treatments. The contextual-bandit and multi-arm bandit approaches provide 72.63% and 64.34% gains, respectively, compared to a random assignment.

Homa Alemzadeh, Ravishankar K. Iyer, Zbigniew Kalbarczyk et al.

Understanding the causes and patient impacts of surgical adverse events will help improve systems and operational practices to avoid incidents in the future. We analyzed the adverse events data related to robotic systems and instruments used in minimally invasive surgery, reported to the U.S. FDA MAUDE database from January 2000 to December 2013. We determined the number of events reported per procedure and per surgical specialty, the most common types of device malfunctions and their impact on patients, and the causes for catastrophic events such as major complications, patient injuries, and deaths. During the study period, 144 deaths (1.4% of the 10,624 reports), 1,391 patient injuries (13.1%), and 8,061 device malfunctions (75.9%) were reported. The numbers of injury and death events per procedure have stayed relatively constant since 2007 (mean = 83.4, 95% CI, 74.2-92.7). Surgical specialties, for which robots are extensively used, such as gynecology and urology, had lower number of injuries, deaths, and conversions per procedure than more complex surgeries, such as cardiothoracic and head and neck (106.3 vs. 232.9, Risk Ratio = 2.2, 95% CI, 1.9-2.6). Device and instrument malfunctions, such as falling of burnt/broken pieces of instruments into the patient (14.7%), electrical arcing of instruments (10.5%), unintended operation of instruments (8.6%), system errors (5%), and video/imaging problems (2.6%), constituted a major part of the reports. Device malfunctions impacted patients in terms of injuries or procedure interruptions. In 1,104 (10.4%) of the events, the procedure was interrupted to restart the system (3.1%), to convert the procedure to non-robotic techniques (7.3%), or to reschedule it to a later time (2.5%). Adoption of advanced techniques in design and operation of robotic surgical systems may reduce these preventable incidents in the future.