A5101841038- Neural Networks and Applications
- Neural dynamics and brain function
- Mitochondrial Function and Pathology
- Bacterial Identification and Susceptibility Testing
- Neural Networks and Reservoir Computing
- Model Reduction and Neural Networks
- Cell death mechanisms and regulation
- Mass Spectrometry Techniques and Applications
- Genomics and Phylogenetic Studies
- Banking stability, regulation, efficiency
- Autophagy in Disease and Therapy
- ATP Synthase and ATPases Research
- Infective Endocarditis Diagnosis and Management
- Plant-Microbe Interactions and Immunity
- Fungal Infections and Studies
- Magnetic and transport properties of perovskites and related materials
- Advanced Memory and Neural Computing
- Advanced Proteomics Techniques and Applications
- Economic Theory and Policy
- Music and Audio Processing
- Theoretical and Computational Physics
- Speech and Audio Processing
- Gaussian Processes and Bayesian Inference
- Blind Source Separation Techniques
- Toxin Mechanisms and Immunotoxins
Johns Hopkins University
2006-2022
Soochow University
2021
University of the Andes
2003-2010
Banco de la República Colombia
2007
Johns Hopkins University Applied Physics Laboratory
1987-2003
University of Maryland, College Park
1997-2001
Massachusetts Institute of Technology
1997
Jet Propulsion Laboratory
1989
Max Planck Institute for Nuclear Physics
1983
An adaptive neural network with asymmetric connections is introduced. This related to the Hopfield graded neurons and uses a recurrent generalization of \ensuremath{\delta} rule Rumelhart, Hinton, Williams modify adaptively synaptic weights. The new bears resemblance master/slave Lapedes Farber but it architecturally simpler.
Mitochondrial fission and fusion are linked to synaptic activity in healthy neurons implicated the regulation of apoptotic cell death many types. We developed fluorescence microscopy computational strategies directly measure mitochondrial frequencies their effects on morphology cultured neurons. found that rate exceeds neuronal processes, and, therefore, fission/fusion ratio alone is insufficient explain at steady state. This imbalance between compensated by growth organelles. Bcl-xL...
Mammalian Bcl-xL protein localizes to the outer mitochondrial membrane, where it inhibits apoptosis by binding Bax and inhibiting Bax-induced membrane permeabilization. Contrary expectation, we found electron microscopy biochemical approaches that endogenous also localized inner cristae. Two-photon of cultured neurons revealed large fluctuations in potential when was genetically deleted or pharmacologically inhibited, indicating increased total ion flux into out mitochondria. Computational,...
Loss or duplication of chromosome segments can lead to further genomic changes associated with cancer. However, it is not known whether only a select subset genes responsible for driving changes. To determine perturbation any given gene in genome suffices drive subsequent genetic changes, we analyzed the yeast knockout collection secondary mutations functional consequence. Unlike wild-type, most strains were found have one additional mutant affecting nutrient responses and/or...
Error backpropagation in feedforward neural network models is a popular learning algorithm that has its roots nonlinear estimation and optimization. It being used routinely to calculate error gradients systems with hundreds of thousands parameters. However, the classical architecture for severe restrictions. The extension networks recurrent connections will be reviewed. now possible efficiently compute have temporal dynamics, which opens applications host problems identification control.
An improved data analysis method is described for rapid identification of intact microorganisms from MALDI-TOF-MS data. The makes no use mass spectral fingerprints. Instead, a microorganism database automatically generated that contains biomarker masses derived ribosomal protein sequences and model N-terminal Met loss. We quantitatively validate the via blind study seeks to identify with known sequences. also include in incompletely sets proteins test specificity method. With an optimal...
We derive and validate a simple statistical model that predicts the distribution of false matches between peaks in matrix-assisted laser desorption/ionization mass spectrometry data proteins proteome databases. The allows us to calculate significance previously reported microorganism identification results. In particular, for Δm = ±1.5 Da, we find computed levels are sufficient demonstrate ability identify microorganisms, provided number candidate microorganisms is limited roughly three...
MALDI-TOF mass spectrometry has been coupled with Internet-based proteome database search algorithms in an approach for direct microorganism identification. This is applied here to characterize intact H. pylori (strain 26695) Gram-negative bacteria, the most ubiquitous human pathogen. A procedure including a specific and common posttranslational modification, N-terminal Met cleavage, algorithm described. Accounting modifications putative protein biomarkers improves identification reliability...
Caspase-dependent apoptotic cell death has been extensively studied in cultured cells and during embryonic development, but the existence of analogous molecular pathways single-cell species is uncertain. This reduced enthusiasm for applying advanced genetic tools available yeast to study regulation. However, partial characterization mammals additional genetically encoded mechanisms, which lead a range dying morphologies necrosis, suggests potential applications genetics. In this light, we...
Unicellular eukaryotes have been suggested as undergoing self-inflicted destruction. However, molecular details are sparse compared with the mechanisms of programmed/regulated cell death known for human cells and animal models. Here, we report a pathway in Saccharomyces cerevisiae leading to vacuole/lysosome membrane permeabilization. Following transient stimulus, yeast die slowly over several hours, consistent an ongoing dying process. A genome-wide screen death-promoting factors identified...
The concept of Contrastive Learning (CL) is developed as a family possible learning algorithms for neural networks. CL an extension Deterministic Boltzmann Machines to more general dynamical systems. During learning, the network oscillates between two phases. One phase has teacher signal and one no signal. weights are updated using rule that corresponds gradient descent on contrast function measures discrepancy free with approach provides unified framework developing new algorithms. It also...
Peer review is fundamentally a cooperative process between scientists in community who agree to each other's work an unbiased fashion. the foundation for decisions concerning publication journals, awarding of grants, and academic promotion. Here we perform laboratory study open closed peer based on online game. We show that when reviewer behavior was made public under review, reviewers were rewarded refereeing formed significantly more interactions (13% increase cooperation, P = 0.018). also...
A representation-independent mean-field dynamics is presented for batched TD(λ). The task learning to predict the outcome of an indirectly observed absorbing Markov process. In case linear representations, discrete-time deterministic iteration affine map whose fixed point can be expressed in closed form without assumption linearly independent observation vectors. Batched TD(λ) proved converge with probability 1 all λ. Theory and simulation agree on a random walk example.