Interpretable deep learning for deconvolutional analysis of neural signals.

The widespread adoption of deep learning to build models that capture the dynamics of neural populations is typically based on "black-box" approaches that lack an interpretable link between neural activity and function. Here, we propose to apply algorithm unrolling, a method for interpretable deep learning, to design the architecture of sparse deconvolutional neural networks and obtain a direct interpretation of network weights in relation to stimulus-driven single-neuron activity through a generative model. We characterize our method, referred to as deconvolutional unrolled neural learning (DUNL), and show its versatility by applying it to deconvolve single-trial local signals across multiple brain areas and recording modalities. To exemplify use cases of our decomposition method, we uncover multiplexed salience and reward prediction error signals from midbrain dopamine neurons in an unbiased manner, perform simultaneous event detection and characterization in somatosensory thalamus recordings, and characterize the responses of neurons in the piriform cortex. Our work leverages the advances in interpretable deep learning to gain a mechanistic understanding of neural dynamics.

Deep Residual Autoencoders for Expectation Maximization-Inspired Dictionary Learning.

We introduce a neural-network architecture, termed the constrained recurrent sparse autoencoder (CRsAE), that solves convolutional dictionary learning problems, thus establishing a link between dictionary learning and neural networks. Specifically, we leverage the interpretation of the alternating-minimization algorithm for dictionary learning as an approximate expectation-maximization algorithm to develop autoencoders that enable the simultaneous training of the dictionary and regularization parameter (ReLU bias). The forward pass of the encoder approximates the sufficient statistics of the E-step as the solution to a sparse coding problem, using an iterative proximal gradient algorithm called FISTA. The encoder can be interpreted either as a recurrent neural network or as a deep residual network, with two-sided ReLU nonlinearities in both cases. The M-step is implemented via a two-stage backpropagation. The first stage relies on a linear decoder applied to the encoder and a norm-squared loss. It parallels the dictionary update step in dictionary learning. The second stage updates the regularization parameter by applying a loss function to the encoder that includes a prior on the parameter motivated by Bayesian statistics. We demonstrate in an image-denoising task that CRsAE learns Gabor-like filters and that the EM-inspired approach for learning biases is superior to the conventional approach. In an application to recordings of electrical activity from the brain, we demonstrate that CRsAE learns realistic spike templates and speeds up the process of identifying spike times by 900× compared with algorithms based on convex optimization.

Robustness of Frequency Division Technique for Online Myoelectric Pattern Recognition against Contraction-Level Variation.

Contraction-level invariant surface electromyography pattern recognition introduces the decrease of training time and decreases the limitation of clinical prostheses. This study intended to examine whether a signal pre-processing method named frequency division technique (FDT) for online myoelectric pattern recognition classification is robust against contraction-level variation, and whether this pre-processing method has an advantage over traditional time-domain pattern recognition techniques even in the absence of muscle contraction-level variation. Eight healthy and naïve subjects performed wrist contractions during two degrees of freedom goal-oriented tasks, divided in three groups of type I, type II, and type III. The performance of these tasks, when the two different methods were used, was quantified by completion rate, completion time, throughput, efficiency, and overshoot. The traditional and the FDT method were compared in four runs, using combinations of normal or high muscle contraction level, and the traditional method or FDT. The results indicated that FDT had an advantage over traditional methods in the tested real-time myoelectric control tasks. FDT had a much better median completion rate of tasks (95%) compared to the traditional method (77.5%) among non-perfect runs, and the variability in FDT was strikingly smaller than the traditional method (p < 0.001). Moreover, the FDT method outperformed the traditional method in case of contraction-level variation between the training and online control phases (p = 0. 005 for throughput in type I tasks with normal contraction level, p = 0.006 for throughput in type II tasks, and p = 0.001 for efficiency with normal contraction level of all task types). This study shows that FDT provides advantages in online myoelectric control as it introduces robustness over contraction-level variations.