Weimer, James
Email Address
ORCID
Disciplines
relationships.isProjectOf
relationships.isOrgUnitOf
Position
Introduction
Research Interests
Collection
23 results
Search Results
Now showing 1 - 10 of 23
Publication Improving Classifier Confidence using Lossy Label-Invariant Transformations(2021-04-01) Jang, Sooyong; Lee, Insup; Weimer, JamesProviding reliable model uncertainty estimates is imperative to enabling robust decision making by autonomous agents and humans alike. While recently there have been significant advances in confidence calibration for trained models, examples with poor calibration persist in most calibrated models. Consequently, multiple techniques have been proposed that leverage label-invariant transformations of the input (i.e., an input manifold) to improve worst-case confidence calibration. However, manifold-based confidence calibration techniques generally do not scale and/or require expensive retraining when applied to models with large input spaces (e.g., ImageNet). In this paper, we present the recursive lossy label-invariant calibration (ReCal) technique that leverages label-invariant transformations of the input that induce a loss of discriminatory information to recursively group (and calibrate) inputs – without requiring model retraining. We show that ReCal outperforms other calibration methods on multiple datasets, especially, on large-scale datasets such as ImageNet.Publication Towards Synthesis of Platform-Aware Attack-Resilient Control Systems: Extended Abstract(2013-04-09) Pajic, Miroslav; Bezzo, Nicola; Weimer, James; Alur, Rajeev; Mangharam, Rahul; Michael, Nathan; Pappas, George J; Sokolsky, Oleg; Tabuada, Paulo; Weirich, Stephanie; Lee, InsupPublication Data-driven Adaptive Safety Monitoring using Virtual Subjects in Medical Cyber-Physical Systems: A Glucose Control Case Study(2016-09-01) Chen, Sanjian; Sokolsky, Oleg; Weimer, James; Lee, InsupMedical cyber-physical systems (MCPS) integrate sensors, actuators, and software to improve patient safety and quality of healthcare. These systems introduce major challenges to safety analysis because the patient’s physiology is complex, nonlinear, unobservable, and uncertain. To cope with the challenge that unidentified physiological parameters may exhibit short-term variances in certain clinical scenarios, we propose a novel run-time predictive safety monitoring technique that leverages a maximal model coupled with online training of a computational virtual subject (CVS) set. The proposed monitor predicts safety-critical events at run-time using only clinically available measurements. We apply the technique to a surgical glucose control case study. Evaluation on retrospective real clinical data shows that the algorithm achieves 96% sensitivity with a low average false alarm rate of 0.5 false alarm per surgery.Publication Attack-Resilient Minimum Mean-Squared Error Estimation(2014-06-01) Weimer, James; Bezzo, Nicola; Pajic, Miroslav; Sokolsky, Oleg; Lee, InsupThis work addresses the design of resilient estimators for stochastic systems. To this end, we introduce a minimum mean-squared error resilient (MMSE-R) estimator whose conditional mean squared error from the state remains finitely bounded and is independent of additive measurement attacks. An implementation of the MMSE-R estimator is presented and is shown as the solution of a semidefinite programming problem, which can be implemented efficiently using convex optimization techniques. The MMSE-R strategy is evaluated against other competing strategies representing other estimation approaches in the presence of small and large measurement attacks. The results indicate that the MMSE-R estimator significantly outperforms (in terms of mean-squared error) other realizable resilient (and non-resilient) estimators.Publication An Intraoperative Glucose Control Benchmark for Formal Verification(2015-10-01) Chen, Sanjian; O'Kelly, Matthew; Weimer, James; Sokolsky, Oleg; Lee, InsupDiabetes associated complications are affecting an increasingly large population of hospitalized patients. Since glucose physiology is significantly impacted by patient-specific parameters, it is critical to verify that a clinical glucose control protocol is safe across a wide patient population. A safe protocol should not drive the glucose level into dangerous low (hypoglycemia) or high (hyperglycemia) ranges. Verification of glucose controllers is challenging due to the high-dimensional, non-linear glucose physiological models which contain both unobservable states and unmeasurable patient-specific parameters. This paper presents a hybrid system model of a closed-loop physiological system that includes an existing FDA-accepted high-fidelity physiological model tailored to intraoperative settings and a validated improvement to a clinical glucose control protocol for diabetic cardiac surgery patients. We propose the closed-loop model as a physiological system benchmark for verification and present our initial results on verifying the system using the SMT-based hybrid system verification tool dReach.Publication Towards Non-Invasive Monitoring of Hypovolemia in Intensive Care Patients(2015-04-13) Roederer, Alexander; Weimer, James; Dimartino, Joseph; Gutsche, Jacob; Lee, InsupHypovolemia caused by internal hemorrhage is a major cause of death in critical care patients. However, hypovolemia is difficult to diagnose in a timely fashion, as obvious symptoms do not manifest until patients are already nearing a critical state of shock. Novel non-invasive methods for detecting hypovolemia in the literature utilize the photoplethysmogram (PPG) waveform generated by the pulse-oximeter attached to a finger or ear. Until now, PPG-based alarms have been evaluated only on healthy patients under ideal testing scenarios (e.g., motionless patients); however, the PPG is sensitive to patient health and significant artifacts manifest when patients move. Since patient health varies within the intensive care unit (ICU) and ICU patients typically do not remain motionless, this work introduces a PPG-based monitor designed to be robust to waveform artifacts and health variability in the underlying patient population. To demonstrate the promise of our approach, we evaluate the proposed monitor on a small sample of intensive care patients from the Physionet database. The monitor detects hypovolemia within a twelve hour window of nurse documentation of hypovolemia when it is present, and achieves a low false alarm rate over patients without documented hypovolemia.Publication Real-Time Detectors for Digital and Physical Adversarial Inputs to Perception Systems(2021-05-01) Kantaros, Yiannis; Carpenter, Taylor; Sridhar, Kaustubh; Yang, Yahan; Lee, Insup; Weimer, JamesDeep neural network (DNN) models have proven to be vulnerable to adversarial digital and physical attacks. In this paper, we propose a novel attack- and dataset-agnostic and real-time detector for both types of adversarial inputs to DNN-based perception systems. In particular, the proposed detector relies on the observation that adversarial images are sensitive to certain label-invariant transformations. Specifically, to determine if an image has been adversarially manipulated, the proposed detector checks if the output of the target classifier on a given input image changes significantly after feeding it a transformed version of the image under investigation. Moreover, we show that the proposed detector is computationally-light both at runtime and design-time which makes it suitable for real-time applications that may also involve large-scale image domains. To highlight this, we demonstrate the efficiency of the proposed detector on ImageNet, a task that is computationally challenging for the majority of relevant defenses, and on physically attacked traffic signs that may be encountered in real-time autonomy applications. Finally, we propose the first adversarial dataset, called AdvNet that includes both clean and physical traffic sign images. Our extensive comparative experiments on the MNIST, CIFAR10, ImageNet, and AdvNet datasets show that VisionGuard outperforms existing defenses in terms of scalability and detection performance. We have also evaluated the proposed detector on field test data obtained on a moving vehicle equipped with a perception-based DNN being under attack.Publication Robust Monitoring of Hypovolemia in Intensive Care Patients Using Photoplethysmogram Signals(2015-08-01) Roederer, Alexander; Weimer, James; Dimartino, Joseph; Gutsche, Jacob; Lee, InsupThe paper presents a fingertip photoplethysmography based technique to assess patient fluid status that is robust to waveform artifacts and health variability in the underlying patient population. The technique is intended for use in intensive care units, where patients are at risk for hypovolemia, and signal artifacts and inter-patient variations in health are common. Input signals are preprocessed to remove artifact, then a parameter-invariant statistic is calculated to remove effects of patient-specific physiology. Patient data from the Physionet MIMICII database was used to evaluate the performance of this technique. The proposed method was able to detect hypovolemia within 24 hours of onset in all hypovolemic patients tested, while producing minimal false alarms over non-hypovolemic patients.Publication Resilient Parameter-Invariant Control With Application to Vehicle Cruise Control(2013-03-20) Weimer, James; Bezzo, Nicola; Pajic, Miroslav; Pappas, George J.; Sokolsky, Oleg; Lee, InsupThis work addresses the general problem of resilient control of unknown stochastic linear time-invariant (LTI) systems in the presence of sensor attacks. Motivated by a vehicle cruise control application, this work considers a first order system with multiple measurements, of which a bounded subset may be corrupted. A frequency-domain-designed resilient parameter-invariant controller is introduced that simultaneously minimizes the effect of corrupted sensors, while maintaining a desired closed-loop performance, invariant to unknown model parameters. Simulated results illustrate that the resilient parameter-invariant controller is capable of stabilizing unknown state disturbances and can perform state trajectory tracking.Publication Towards Assurance Cases for Resilient Control Systems(2014-08-01) Weimer, James; Sokolsky, Oleg; Bezzo, Nicola; Lee, InsupThe paper studies the problem of constructing assurance cases for embedded control systems developed using a model-based approach. Assurance cases aim to provide a convincing argument that the system delivers certain guarantees, based on the evidence obtained during the design and evaluation of the system. We suggest an argument strategy centered around properties of models used in the development and properties of tools that manipulate these models. The paper presents the case study of a resilient speed estimator for an autonomous ground vehicle and takes the reader through a detailed assurance case arguing that the estimator computes speed estimates with bounded error.