A5011183160- Target Tracking and Data Fusion in Sensor Networks
- Distributed Control Multi-Agent Systems
- Adversarial Robustness in Machine Learning
- Distributed Sensor Networks and Detection Algorithms
- Electric Motor Design and Analysis
- Reinforcement Learning in Robotics
- Stochastic Gradient Optimization Techniques
- Topic Modeling
- Anomaly Detection Techniques and Applications
- Gaussian Processes and Bayesian Inference
- Advanced Bandit Algorithms Research
- Explainable Artificial Intelligence (XAI)
- Advanced Control Systems Optimization
- Time Series Analysis and Forecasting
- Machine Learning in Healthcare
- Innovative Energy Harvesting Technologies
- Generative Adversarial Networks and Image Synthesis
- Domain Adaptation and Few-Shot Learning
- Control Systems and Identification
- Fault Detection and Control Systems
- Bayesian Methods and Mixture Models
- Chronic Disease Management Strategies
- Online Learning and Analytics
- Machine Learning and ELM
- Neural Networks and Applications
IBM (United States)
2017-2025
Cambridge Scientific (United States)
2020-2025
Indian Institute of Technology Kanpur
2021
IBM Research - Thomas J. Watson Research Center
2016-2017
Carnegie Mellon University
2013-2015
Laboratoire de Traitement de l'Information Médicale
2012
Massachusetts Institute of Technology
2005-2006
This paper studies distributed estimation of unstable dynamic random fields observed by a sparsely connected network sensors. The field dynamics are globally detectable, but not necessarily locally detectable. We propose consensus+innovations estimator, termed Distributed Information Kalman Filter. prove under what conditions this estimator is asymptotically unbiased with bounded mean-squared error, smaller than for other alternative estimators. Monte Carlo simulations confirm our...
In this paper, we address the distributed filtering and prediction of time-varying random fields represented by linear time-invariant (LTI) dynamical systems. The field is observed a sparsely connected network agents/sensors collaborating among themselves. We develop Kalman filter type consensus + innovations estimator dynamic termed as Consensus+Innovations Filter. analyze convergence properties estimator. prove that mean-squared error asymptotically converges if degree instability dynamics...
Concept bottleneck models (CBM) are a popular way of creating more interpretable neural networks by having hidden layer neurons correspond to human-understandable concepts. However, existing CBMs and their variants have two crucial limitations: first, they need collect labeled data for each the predefined concepts, which is time consuming labor intensive; second, accuracy CBM often significantly lower than that standard network, especially on complex datasets. This poor performance creates...
Large language models (LLMs) have demonstrated remarkable reasoning capabilities across diverse domains. Recent studies shown that increasing test-time computation enhances LLMs' capabilities. This typically involves extensive sampling at inference time guided by an external LLM verifier, resulting in a two-player system. Despite guidance, the effectiveness of this system demonstrates potential single to tackle complex tasks. Thus, we pose new research problem: Can internalize searching...
We introduce Granite Vision, a lightweight large language model with vision capabilities, specifically designed to excel in enterprise use cases, particularly visual document understanding. Our is trained on comprehensive instruction-following dataset, including document-related tasks, such as content extraction from tables, charts, diagrams, sketches, and infographics, well general image tasks. The architecture of Vision centered around modality alignment decoder-only, 2 billion parameter...
The recent wave of AI and automation has been argued to differ from previous General Purpose Technologies (GPTs), in that it may lead rapid change occupations' underlying task requirements persistent technological unemployment. In this paper, we apply a novel methodology dynamic shares large dataset online job postings explore how exactly occupational demands have changed over the past decade innovation, especially across high, mid low wage occupations. Notably, big data risen significantly...
Deploying deep reinforcement learning in safety-critical settings requires developing algorithms that obey hard constraints during exploration. This paper contributes a first approach toward enforcing formal safety on end-to-end policies with visual inputs. Our draws recent advances object detection and automated reasoning for hybrid dynamical systems. The is evaluated novel benchmark emphasizes the challenge of safely exploring presence constraints. from several proposed problem sets safe...
.Generative adversarial networks (GANs) learn the distribution of observed samples through a zero-sum game between two machine players, generator and discriminator. While GANs achieve great success in learning complex image, sound, text data, they perform suboptimally multimodal distribution-learning benchmarks such as Gaussian mixture models (GMMs). In this paper, we propose Generative Adversarial Training for Mixture Models (GAT-GMM), minimax GAN framework GMMs. Motivated by optimal...
It is known that neural networks have the problem of being over-confident when directly using output label distribution to generate uncertainty measures. Existing methods mainly resolve this issue by retraining entire model impose quantification capability so learned can achieve desired performance in accuracy and prediction simultaneously. However, training from scratch computationally expensive, a trade-off might exist between quantification. To end, we consider more practical post-hoc...
Deep learning, the most important subfield of machine learning and artificial intelligence (AI) over last decade, is considered one fundamental technologies underpinning Fourth Industrial Revolution. But despite its record-breaking history, deep learning’s enormous appetite for compute data means that sometimes it can be too costly to practically use. In this paper, we connect technical insights from scaling laws transfer with economics IT propose a framework estimating cost computer vision...
This paper presents the design, fabrication, and characterization of permanent-magnet (PM) generators for use in microscale power generation systems. The are three-phase, axial-flux, synchronous machines, each consisting an eight-pole surface-wound stator PM rotor. devices fabricated using a combination microfabrication precision assembly. Characterization is achieved custom-built experimental test stand that incorporates off-the-shelf gas-driven spindle to machines. At rotational speed 120...
In this paper, we consider the problem of state estimation a dynamical system in multi-agent network. The agents are sparsely connected and each them observes strict subset vector. distributed algorithm that propose enables agent to estimate any arbitrary linear with bounded mean-squared error. To achieve this, ratio algebraic connectivity largest eigenvalue graph Laplacian has be larger than lower bound determined by spectral radius system's dynamics matrix. This extends notion Network...
The dearth of prescribing guidelines for physicians is one key driver the current opioid epidemic in United States. In this work, we analyze medical and pharmaceutical claims data to draw insights on characteristics patients who are more prone adverse outcomes after an initial synthetic prescription. Toward end, propose a generative model that allows discovery from observational subgroups demonstrate enhanced or diminished causal effect due treatment. Our approach models these...
This paper presents the modeling of permanent-magnet (PM) generators for use in microscale power generation systems. The are three-phase, axial-flux, synchronous machines, each consisting a multipole, surface-wound stator and PM rotor. machines modeled by analytically solving two-dimensional (2-D) magneto-quasi-static Maxwell's equations as function radius. 2-D field solutions then integrated over radial span machine to determine circuit parameters such open-circuit voltage inductance well...
This paper considers online optimal control with affine constraints on the states and actions under linear dynamics bounded random disturbances. The system are assumed to be known time invariant but convex stage cost functions change adversarially. To solve this problem, we propose Online Gradient Descent Buffer Zones (OGD-BZ). Theoretically, show that OGD-BZ proper parameters can guarantee satisfy all despite any admissible Further, investigate policy regret of OGD-BZ, which compares...
This paper presents the design, fabrication, and characterization of laminated, magnetic induction machines intended for high-speed, high-temperature, high-power-density, silicon-based microengine power generation systems. Innovative fabrication techniques were used to embed electroplated materials (Cu, Ni/sub 80/Fe/sub 20/, Co/sub 65/Fe/sub 18/Ni/sub 17/) within bulk-micromachined fusion-bonded silicon form machine structures. The characterized in motoring mode using tethered rotors,...
In recent years, there has been growing interest in the use of fitness trackers and smartphone applications for promoting physical activity. Many these accelerometers to estimate level activity that users engage provide visual reports a user's step counts. When provided, most recommendations are limited popular general health advice. our study, we develop an approach providing data-driven personalized intraday planning. We generate hour-by-hour plan is based on probability adhering plan. The...
This paper explores a modern predictive uncertainty estimation approach, called evidential deep learning (EDL), in which single neural network model is trained to learn meta distribution over the by minimizing specific objective function. Despite their strong empirical performance, recent studies Bengs et al. identify fundamental pitfall of existing methods: learned epistemic may not vanish even infinite-sample limit. We corroborate observation providing unifying view class widely used...
Neural networks are powerful tools in various applications, and quantifying their uncertainty is crucial for reliable decision-making. In the deep learning field, uncertainties usually categorized into aleatoric (data) epistemic (model) uncertainty. this paper, we point out that existing popular variance attenuation method highly overestimates To address issue, proposed a new estimation by actively de-noising observed data. By conducting broad range of experiments, demonstrate our approach...
We study the adaptive control of an unknown linear system with a quadratic cost function subject to safety constraints on both states and actions. The challenges this problem arise from tension among safety, exploration, performance, computation. To address these challenges, we propose polynomial-time algorithm that guarantees feasibility constraint satisfaction high probability under proper conditions. Our is implemented single trajectory does not require restarts. Further, analyze regret...
Due to privacy or commercial constraints, large pre-trained language models (PLMs) are often offered as black-box APIs. Fine-tuning such downstream tasks is challenging because one can neither access the model's internal representations nor propagate gradients through it. This paper addresses these challenges by developing techniques for adapting PLMs with only API access. Building on recent work soft prompt tuning, we develop methods tune prompts without requiring gradient computation....
This paper presents the design, fabrication, and characterization of permanent magnet (PM) generators for use in microscale power generation systems. The are three phase, axial flux, synchronous machines, each consisting a multipole surface wound stator PM rotor. microfabricated windings, with small interconductor gaps variable width geometry, key enabler high density. At rotational speed 120,000 rpm, one such generator demonstrates 2.6 W mechanical-to-electrical conversion and, coupled to...