arXiv subjects

Jose C Principe

Publications and source records attributed to Jose C Principe.

4 recordsLinked to original sources

arXiv2026

A Stable Neural Statistical Dependence Estimator for Autoencoder Feature Analysis

Statistical dependence measures like mutual information is ideal for analyzing autoencoders, but it can be ill-posed for deterministic, static, noise-free networks. We adopt the variational (Gaussian) formulation that makes dependence among inputs, latents, and reconstructions measurable, and we propose a stable neural dependence estimator based on an orthonormal density-ratio decomposition. Unlike MINE, our method avoids input concatenation and product-of-marginals re-pairing, reducing computational cost and improving stability. We introduce an efficient NMF-like scalar cost and demonstrate empirically that assuming Gaussian noise to form an auxiliary variable enables meaningful dependence measurements and supports quantitative feature analysis, with a sequential convergence of singular values.

cs.LG↗

arXiv2019

Unsupervised decoding of spinal motor neuron spike trains for estimating hand kinematics following targeted muscle reinnervation

The performance of upper-limb prostheses is currently limited by the relatively poor functionality of unintuitive control schemes. This paper proposes to extract, from multichannel electromyographic signals (EMG), motor neuron spike trains and project them into lower dimensional continuous signals, which are used as multichannel proportional inputs to control the prosthetic's actuators. These control signals are an estimation of the common synaptic input that the motor neurons receive. We use the simplest of metric learning approaches known as principal component analysis (PCA), as a linear unsupervised metric projection to extract the spectral information of high dimensional data into the subspace of the prosthetic hand degrees of freedom. We also investigate the importance of a rotation in the projection space that best aligns the PCA subspace with the space of degrees of freedom of the prosthetic hand, to attempt to approximate the sparseness properties of the motor axes (which are orthogonal), while no sparseness is enforced by PCA. Proof of concept for the feedforward path (open loop) is given by successful estimation of concurrent movements with up to three degrees of freedom. We also analyze and quantify the performance of the proposed decoding algorithm under implementation constraints and hardware limitations and propose a space-time subsampling strategy, to maximize projection fidelity, in each extracted source over time. The results confirm that the proposed decoding approach can directly project the motor neuron spike trains into kinematics for simultaneous and proportional prosthesis control that can be extended to multiple degrees of freedom. We show that the method is robust to reducing the training data in space (number of sources) and time, which makes it potentially suitable for clinical applications.

eess.SP↗

arXiv2017

Exploiting Spatio-Temporal Structure with Recurrent Winner-Take-All Networks

We propose a convolutional recurrent neural network, with Winner-Take-All dropout for high dimensional unsupervised feature learning in multi-dimensional time series. We apply the proposedmethod for object recognition with temporal context in videos and obtain better results than comparable methods in the literature, including the Deep Predictive Coding Networks previously proposed by Chalasani and Principe.Our contributions can be summarized as a scalable reinterpretation of the Deep Predictive Coding Networks trained end-to-end with backpropagation through time, an extension of the previously proposed Winner-Take-All Autoencoders to sequences in time, and a new technique for initializing and regularizing convolutional-recurrent neural networks.

cs.LG↗

arXiv2016

Information Theoretic-Learning Auto-Encoder

We propose Information Theoretic-Learning (ITL) divergence measures for variational regularization of neural networks. We also explore ITL-regularized autoencoders as an alternative to variational autoencoding bayes, adversarial autoencoders and generative adversarial networks for randomly generating sample data without explicitly defining a partition function. This paper also formalizes, generative moment matching networks under the ITL framework.

cs.LG↗