A5025310174- Topic Modeling
- Natural Language Processing Techniques
- Functional Brain Connectivity Studies
- Sparse and Compressive Sensing Techniques
- Distributed systems and fault tolerance
- Neural Networks and Applications
- Text Readability and Simplification
- Face and Expression Recognition
- Statistical Methods and Inference
- Evolutionary Algorithms and Applications
- Parallel Computing and Optimization Techniques
- Image and Signal Denoising Methods
- Advanced Database Systems and Queries
- Advanced Neuroimaging Techniques and Applications
- Advanced Data Storage Technologies
- Optimization and Search Problems
- Machine Learning and Data Classification
- Multimodal Machine Learning Applications
- Advanced MRI Techniques and Applications
- Software Testing and Debugging Techniques
- Photoacoustic and Ultrasonic Imaging
- Software Reliability and Analysis Research
- Domain Adaptation and Few-Shot Learning
- 3D Shape Modeling and Analysis
- Electrical and Bioimpedance Tomography
Intel (United States)
2016-2023
Intelligent Systems Research (United States)
2023
The University of Texas at Austin
2021-2023
Shanghai Center for Brain Science and Brain-Inspired Technology
2019
Intel (Germany)
2019
Technion – Israel Institute of Technology
2011-2016
IBM (United States)
1990-2009
IBM Research - Haifa
2008
IBM Research - Thomas J. Watson Research Center
1993-2002
Courant Institute of Mathematical Sciences
2002
A model for studying the optimal allocation of cache memory among two or more competing processes is developed and used to show that, examples studied, least recently (LRU) replacement strategy produces allocations that are very close optimal. It also shown when program behavior changes, LRU moves quickly toward steady-state if it far from optimal, but converges slowly as approaches allocation. An efficient combinatorial algorithm determining allocation, which, in theory, could be reduce...
Known results regarding consensus among processors are surveyed and related to practice. The ideas embodied in the various proofs explained. goal is give practitioners some sense of system hardware software guarantees that required achieve a given level reliability performance. survey focuses on two categories failures: fail-stop failures, which occur when fail by stopping; Byzantine acting maliciously.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML"...
Functional magnetic resonance imaging (fMRI) offers a rich source of data for studying the neural basis cognition. Here, we describe Brain Imaging Analysis Kit (BrainIAK), an open-source, free Python package that provides computationally optimized solutions to key problems in advanced fMRI analysis. A variety techniques are presently included BrainIAK: intersubject correlation (ISC) and functional connectivity (ISFC), alignment via shared response model (SRM), full matrix analysis (FCMA),...
A multiple reservation approach that allows atomic updates of shared variables and simplifies concurrent nonblocking codes for managing data structures such as queues linked lists is presented. The method can be implemented an extension to any cache protocol grants write access at most one processor a time. Performance improvement, automatic restart, livelock avoidance are discussed. Some sample programs examined.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML"...
Functional magnetic resonance imaging (fMRI) offers a rich source of data for studying the neural basis cognition. Here, we describe Brain Imaging Analysis Kit (BrainIAK), an open-source, free Python package that provides computationally-optimized solutions to key problems in advanced fMRI analysis. A variety techniques are presently included BrainIAK: intersubject correlation (ISC) and functional connectivity (ISFC), alignment via shared response model (SRM), full matrix analysis (FCMA),...
In ultrasound imaging, clutter artifacts degrade images and may cause inaccurate diagnosis. this paper, we apply a method called Morphological Component Analysis (MCA) for sparse signal separation with the objective of reducing such artifacts. The MCA approach assumes that two signals in additive mix have each representation under some dictionary atoms (a matrix), is achieved by finding these representations. our work, an adaptive used learning from echo data. compared to Singular Value...
As Large Language Models (LLMs) evolve from text-completion tools into fully fledged agents operating in dynamic environments, they must address the challenge of continually learning and retaining long-term knowledge. Many biological systems solve these challenges with episodic memory, which supports single-shot instance-specific contexts. Inspired by this, we present an memory framework for LLM agents, centered around five key properties that underlie adaptive context-sensitive behavior....
Presents a parallel hash join algorithm that is based on the concept of hierarchical hashing, to address problem data skew. The proposed splits usual phase into and an explicit transfer phase, adds extra scheduling between these two. During heuristic optimization algorithm, using output attempts balance load across multiple processors in subsequent phase. naturally identifies partitions with largest skew values them as necessary, assigning each optimal number processors. Assuming for...
Among the many ways to model signals, a recent approach that draws considerable attention is sparse representation modeling.In this model, signal assumed be generated as random linear combination of few atoms from pre-specified dictionary.In work we analyze two Bayesian denoising algorithms -the Maximum-Aposteriori Probability (MAP) and Minimum-Mean-Squared-Error (MMSE) estimators, under assumption dictionary unitary.It well known both these estimators lead scalar shrinkage on transformed...
The scale of functional magnetic resonance image data is rapidly increasing as large multi-subject datasets are becoming widely available and high-resolution scanners adopted. inherent low-dimensionality the information in this has led neuroscientists to consider factor analysis methods extract analyze underlying brain activity. In work, we two recent methods: Shared Response Model Hierarchical Topographic Factor Analysis. We perform analytical, algorithmic, code optimization enable...
Abstract Natural language contains information at multiple timescales. To understand how the human brain represents this information, one approach is to build encoding models that predict fMRI responses natural using representations extracted from neural network (LMs). However, these LM-derived do not explicitly separate different timescales, making it difficult interpret models. In work we construct interpretable multi-timescale by forcing individual units in an LSTM LM integrate over...
Cognitive neuroscience seeks to explain the organization of brain, but typically focuses on aspects that are shared across people rather than those vary individuals. Here, we present a new method for analyzing brain imaging data captures both and individual components activity. Inspired by response model (SRM) robust principal analysis, (RSRM) aligns functional topographies humans while preserving component sparse, Experimental results adult showed RSRM performs as well or better SRM, at...
Finding the most effective way to aggregate multi-subject fMRI data is a long-standing and challenging problem. It of increasing interest in contemporary studies human cognition due scarcity per subject variability brain anatomy functional response across subjects. Recent work on latent factor models shows promising results this task but approach does not preserve spatial locality brain. We examine two ways combine ideas model searchlight based analysis while preserving locality. first do...
How related are the representations learned by neural language models, translation and tagging tasks? We answer this question adapting an encoder-decoder transfer learning method from computer vision to investigate structure among 100 different feature spaces extracted hidden of various networks trained on tasks. This reveals a low-dimensional where models smoothly interpolate between word embeddings, syntactic semantic tasks, future embeddings. call representation embedding because it...
MRI is commonly used to evaluate pediatric musculoskeletal pathologies, but same-day/near-term scheduling and short exams remain challenges.To investigate the feasibility of a targeted rapid knee exam, with goal reducing cost enabling same-day access.A effectiveness study done prospectively.Forty-seven patients.3T. The 10-minute protocol was based on T2 Shuffling, four-dimensional acquisition reconstruction images variable contrast, T1 2D fast spin-echo (FSE) sequence. A distributed,...
Practical limitations on the duration of individual fMRI scans have led neuroscientist to consider aggregation data from multiple subjects. Differences in anatomical structures and functional topographies brains require aligning across Existing alignment methods serve as a preprocessing step that allows subsequent statistical learn aggregated multi-subject data. Despite their success, current do not leverage labeled used methods. In this work we propose semi-supervised scheme simultaneously...
Solving $l_1$ regularized optimization problems is common in the fields of computational biology, signal processing, and machine learning. Such regularization utilized to find sparse minimizers convex functions. A well-known example least absolute shrinkage selection operator (LASSO) problem, where norm regularizes a quadratic function. multilevel framework presented for solving such efficiently. We take advantage expected sparseness solution, create hierarchy similar type, which traversed...
Language models must capture statistical dependencies between words at timescales ranging from very short to long. Earlier work has demonstrated that in natural language tend decay with distance according a power law. However, it is unclear how this knowledge can be used for analyzing or designing neural network models. In work, we derived theory the memory gating mechanism long short-term (LSTM) law decay. We found unit within an LSTM, which are determined by forget gate bias, should follow...
Software configurations play a crucial role in determining the behavior of software systems. In order to ensure safe and error-free operation, it is necessary identify correct configuration, along with their valid bounds rules, which are commonly referred as specifications. As systems grow complexity scale, number associated specifications required operation can become large prohibitively difficult manipulate manually. Due fast pace development, often case that not thoroughly checked or...
Parallel processing is an attractive option for relational database systems. As in any parallel environment however, load balancing a critical issue which affects overall performance. Load one common operation particular, the join of two relations, can be severely hampered conventional algorithms, due to natural phenomenon known as data skew. In pair recent papers (J. Wolf et al., 1993; 1993), we described new algorithms designed address skew problem. We propose significant improvements both...
We introduce a new framework for classifying large images (in the EOS; Earth Observing System) that is more accurate and less computationally expensive than classical pixel-by-pixel approach. This approach, called progressive classification, well suited analyzing images, such as multispectral satellite scenes, compressed with wavelet-based or block-transform-based transformations. These transformations produce multiresolution pyramid representation of data. A classifier analyses image at...