Closed-Loop Neural Interfaces with Embedded Machine Learning

Uisub Shin, Mahsa Shoaran, Bingzhao Zhu
IEEE, 2020
Conference paper

Abstract

Neural interfaces capable of multi-site electrical recording, on-site signal classification, and closed-loop therapy are critical for the diagnosis and treatment of neurological disorders. However, deploying machine learning algorithms on low-power neural devices is challenging, given the tight constraints on computational and memory resources for such devices. In this paper, we review the recent developments in embedding machine learning in neural interfaces, with a focus on design trade-offs and hardware efficiency. We also present our optimized tree-based model for low-power and memory-efficient classification of neural signal in brain implants. Using energy-aware learning and model compression, we show that the proposed oblique trees can outperform conventional machine learning models in applications such as seizure or tremor detection and motor decoding.

Official source

https://infoscience.epfl.ch/record/284224?ln=en

About this result

This page is automatically generated and may contain information that is not correct, complete, up-to-date, or relevant to your search query. The same applies to every other page on this website. Please make sure to verify the information with EPFL's official sources.

Related concepts

Related publications

Evolutionary Advantages of Neuromodulated Plasticity in Dynamic, Reward- based Scenarios

Peter Dürr, Dario Floreano, Claudio Mattiussi, Andrea Soltoggio

memory in biological neural networks. Similarly, artificial neural networks could benefit from modulatory dynamics when facing certain types of learning problem. Here we test this hypothesis by introducing modulatory neurons to enhance or dampen neural plasticity at target neural nodes. Simulated evolution is employed to design neural control networks for T-maze learning problems, using both standard and modulatory neurons. The results show that experiments where modulatory neurons are enabled achieve better learning in comparison to those where modulatory neurons are disabled. We conclude that modulatory neurons evolve autonomously in the proposed learning tasks, allowing for increased learning and memory capabilities.

MIT Press2008

Constrained Binary Identification Problem

Amin Karbasi, Morteza Zadimoghaddam

We consider the problem of building a binary decision tree, to locate an object within a set by way of the least number of membership queries. This problem is equivalent to the “20 questions game” of information theory and is closely related to lossless source compression. If any query is admissible, Huffman coding is optimal with close to H[P] questions on average, the entropy of the prior distribution P over objects. However, in many realistic scenarios, there are constraints on which queries can be asked, and solving the problem optimally is NP-hard. We provide novel polynomial time approximation algorithms where constraints are defined in terms of “graph", general “cost", and “submodular" functions. In particular, we show that under graph constraints, there exists a constant approximation algorithm for locating the target in the set. We then extend our approach for scenarios where the constraints are defined in terms of general cost functions that depend only on the size of the query and provide an approxima- tion algorithm that can find the target within O(log(log n)) gap from the cost of the optimum algorithm. Submodular functions come as a natural generalization of cost functions with decreas- ing marginals. Under submodular set constraints, we devise an approximation algorithm that can find the target within O(log n) gap from the cost of the optimum algorithm. The proposed algorithms are greedy in a sense that at each step they select a query that most evenly splits the set without violating the underlying constraints. These results can be applied to network tomography, active learning and interactive content search.

2013

Smart and miniaturized implantable microsystems with diagnostic and therapeutic capabilities are becoming increasingly important for patients suffering from neurological disorders such as epilepsy. Recent developments in microfabrication technology have provided new insights into seizure generation at an unprecedented spatial scale. Based on these findings, designing powerful acquisition systems capable of probing the wide-range spatiotemporal activities within the brain holds a great promise to improve the quality of life of epileptic patients. It is expected that future neural recording interfaces will integrate hundreds of acquisition channels at relatively high sampling rates, making it crucial to perform sophisticated signal conditioning tasks within the sensors. A small size of the implantable system is critical to minimize potential clinical issues associated with implantation. The total power consumption should be minimized to avoid heat generation inside the brain. Substantial improvements in neural implant safety and longevity are needed to translate existing multisite neural recording systems into technology suitable for long-term use in patients. As a major technological barrier, the high overall data rates of digitized neural signals collected by dense electrode arrays can drastically increase the power consumption of the wireless transmission module. Consequently, extensive system-level design improvement is needed to meet the requirements of the implantable device, while preserving the high-resolution monitoring capability. In this context, we target the exploration and development of novel methods and tools for more accurate monitoring of the electrical activity of the epileptic brain during seizures. A precise focus localization and accurate early seizure detection are two major goals. Low-power circuit and system design techniques for data acquisition, compression and seizure detection in multichannel cortical implants are presented. In this context, the first fully-integrated circuit that addresses the multichannel compressed-domain feature extraction is proposed. Circuit challenges and techniques to optimize the neural amplifier topology at the frontend of the signal acquisition chain, in terms of power consumption, die area and noise are discussed. Then, compressive sensing is utilized as the main data reduction method in the proposed system. The existing microelectronic implementations of compressive sensing are applied in a single-channel basis. Therefore, these topologies incur a high power consumption and large silicon area. To overcome these issues, a new multichannel measurement scheme and an appropriate recovery scheme are proposed, which encode the whole array into a single compressed data stream. This technique circumvents the need to place one analog-to-digital converter inside each channel, which results in a significant area saving. Taking benefit of the area-efficient implementation of compressive sensing, the number of recording units implantable on the cortex which satisfy the energy constraints of the system is scaled up by a factor equal to the compression ratio. Finally, a new feature extraction method applied on the compressed signal of several channels is proposed. The spatial evolution of the intracranial EEG signals recorded by the adjacent electrodes and the corresponding behaviour in the compressed domain are exploited for detecting seizure onset.

EPFL2015