A study of autoencoders as a feature extraction technique for spike sorting

Abstract

Spike sorting is the process of grouping spikes of distinct neurons into their respective clusters. Most frequently, this grouping is performed by relying on the similarity of features extracted from spike shapes. In spite of recent developments, current methods have yet to achieve satisfactory performance and many investigators favour sorting manually, even though it is an intensive undertaking that requires prolonged allotments of time. To automate the process, a diverse array of machine learning techniques has been applied. The performance of these techniques depends however critically on the feature extraction step. Here, we propose deep learning using autoencoders as a feature extraction method and evaluate extensively the performance of multiple designs. The models presented are evaluated on publicly available synthetic and real “in vivo” datasets, with various numbers of clusters. The proposed methods indicate a higher performance for the process of spike sorting when compared to other state-of-the-art techniques.

Intelligent intrusion detection through deep autoencoder and stacked long short-term memory

“…One of the key reasons for utilizing the autoencoder for feature extraction in the UNSW-NB15 dataset is its capacity to reduce data dimensionality [20]. By learning to represent data succinctly, the autoencoder enables the compression of information while retaining crucial features for intrusion detection.…”

“…The 10 decoding layers perform the inverse operations of the encoding layers to reconstruct the data from the latent code. For each decoding layer 𝑘 ∈ [11,20], the computations are defined by (4).…”

“…The final output layer 𝑦 ^ produces the ultimate reconstruction of the input data from the output of the last decoding layer ℎ (20) . For the training of the deep autoencoder, a loss function is employed to evaluate the disparity between the input data 𝑥 and the reconstructed output data 𝑦 ^.…”

Towards robust and generalizable representations of extracellular data using contrastive learning

“…Despite the ubiquity of PCA as a feature extraction method for spike sorting, it lacks robustness to common nuisance variables in extracellular recordings, cannot capture non-linear features of the observed data, and prefers features that explain variance rather than those that discriminate different waveforms. More recently, it has been shown that non-linear autoencoders can have higher performance than PCA despite suffering from similar drawbacks [44].…”

Clustering of LMS Use Strategies with Autoencoders

“…In addition, variational autoencoders are used to learn a latent data representation that captures the natural clustering of bank customers' data according to creditworthiness [19]. In addition, Typical applications of autoencoders include dimensionality reduction [13] and feature extraction [12,14]. Traditional ways to reduce the dimensionality of large datasets are removing variables with a high correlation and/or a high number of missing values, as well as using PCA (Principal Components Analysis).…”