Temporal sequences are an important feature of neural information processing in biology. Neurons can fire a spike with millisecond precision, and, at the network level, repetitions of spatiotemporal spike patterns are observed in neurobiological data. However, methods for detecting precise temporal patterns in neural activity suffer from high computational complexity and poor robustness to noise, and quantitative detection of these repetitive patterns remains an open problem. Here, we propose a new method to extract spike patterns embedded in raster plots using a 1D convolutional autoencoder with the Earth Mover’s Distance (EMD) as a loss function. Importantly, the properties of the EMD make the method suitable for spike-based distributions, easy to compute, and robust to noise. Through gradient descent, the autoencoder is trained to minimize the EMD between the input and its reconstruction. We then expect the weight matrices to learn the repeating spike patterns present in the data. We validate our method on synthetically generated raster plots and compare its performance with an autoencoder trained using the Mean Squared Error (MSE) as a loss function. We show that the method using the EMD performs better at detecting the occurrence of the spike patterns, while the method using the MSE is better at capturing the underlying distributions used to generate the spikes. Finally, we propose to train the autoencoder iteratively by sequentially combining the EMD and the MSE losses. This sequential approach outperforms the widely used seqNMF method in terms of robustness to various types of noise, speed and stability. Overall, our method provides a novel approach to reliably extract repetitive temporal spike sequences, and can be readily generalized to other sequence detection applications.

Accurate time-series forecasting is essential across a multitude of scientific and industrial domains, yet deep learning models often struggle with challenges such as capturing long-term dependencies and adapting to drift in data distributions over time. We introduce Future-Guided Learning, an approach that enhances time-series event forecasting through a dynamic feedback mechanism inspired by predictive coding. Our approach involves two models: a detection model that analyzes future data to identify critical events and a forecasting model that predicts these events based on present data. When discrepancies arise between the forecasting and detection models, the forecasting model undergoes more substantial updates, effectively minimizing surprise and adapting to shifts in the data distribution by aligning its predictions with actual future outcomes. This feedback loop, drawing upon principles of predictive coding, enables the forecasting model to dynamically adjust its parameters, improving accuracy by focusing on features that remain relevant despite changes in the underlying data. We validate our method on a variety of tasks such as seizure prediction in biomedical signal analysis and forecasting in dynamical systems, achieving a 40% increase in the area under the receiver operating characteristic curve (AUC-ROC) and a 10% reduction in mean absolute error (MAE), respectively. By incorporating a predictive feedback mechanism that adapts to data distribution drift, Future-Guided Learning offers a promising avenue for advancing time-series forecasting with deep learning.

We propose a neuromimetic architecture able to perform always-on pattern recognition. To achieve this, we extended an existing event-based algorithm [1], which introduced novel spatio-temporal features as a Hierarchy Of Time-Surfaces (HOTS). Built from asynchronous events acquired by a neuromorphic camera, these time surfaces allow to code the local dynamics of a visual scene and to create an efficient event-based pattern recognition architecture. Inspired by neuroscience, we extended this method to increase its performance. Our first contribution was to add a homeostatic gain control on the activity of neurons to improve the learning of spatio-temporal patterns [2]. A second contribution is to draw an analogy between the HOTS algorithm and Spiking Neural Networks (SNN). Following that analogy, our last contribution is to modify the classification layer and remodel the offline pattern categorization method previously used into an online and event-driven one. This classifier uses the spiking output of the network to define novel time surfaces and we then perform online classification with a neuromimetic implementation of a multinomial logistic regression. Not only do these improvements increase consistently the performances of the network, they also make this event-driven pattern recognition algorithm online and bio-realistic. Results were validated on different datasets: DVS barrel [3], Poker-DVS [4] and N-MNIST [5]. We foresee to develop the SNN version of the method and to extend this fully event-driven approach to more naturalistic tasks, notably for always-on, ultra-fast object categorization.

Motion Clouds are a generative model for naturalistic visual stimulation that offer full parametric control and more naturalism than the widely used alternatives of Random Dot Kinematograms (RDKs) or luminance gratings. We previously released an 3D FFT-based generation algorithm (Sanz-Leon et al., J Neurophysiol, 2012). Here, we present a novel implementation of motion clouds that uses an Auto-Regressive formulation so that any number of frames can be generated quickly with parameters changed in near real time, as needed in closed loop experiments. We demonstrate a version of the proposed toolbox that will be available online to illustrate the level of control available. With a graphic user interface, researchers can use interactive sliders to adjust motion cloud parameters like central frequency, orientations and bandwidths to get an intuitive feel for the parametric changes. We provide functions that can be easily integrated with psychophysics task tools like Psychtoolbox. Motion clouds can be used to generate trials of stand-alone moving luminance textures or added to other stimuli like images or videos as dynamic noise to disrupt visual processing. The toolbox can be run using GPUs to speed up generation to pseudo real-time for large stimulus arrays of about 1024 by 1024 pixels at 100Hz. We argue that this tool can enhance visual perception experiments in a range of contexts and would like it to be open to extensive testing, use and further development by the psychophysics, computational modelling, functional imaging and neurophysiology communities.

Both biological and artificial neural networks inherently balance their performance with their operational cost, which balances their computational abilities. Typically, an efficient neuromorphic neural network is one that learns representations that reduce the redundancies and dimensionality of its input. This is for instance achieved in sparse coding, and sparse representations derived from natural images yield representations that are heterogeneous, both in their sampling of input features and in the variance of those features. Here, we investigated the connection between natural images’ structure, particularly oriented features, and their corresponding sparse codes. We showed that representations of input features scattered across multiple levels of variance substantially improve the sparseness and resilience of sparse codes, at the cost of reconstruction performance. This echoes the structure of the model’s input, allowing to account for the heterogeneously aleatoric structures of natural images. We demonstrate that learning kernel from natural images produces heterogeneity by balancing between approximate and dense representations, which improves all reconstruction metrics. Using a parametrized control of the kernels’ heterogeneity used by a convolutional sparse coding algorithm, we show that heterogeneity emphasizes sparseness, while homogeneity improves representation granularity. In a broader context, these encoding strategy can serve as inputs to deep convolutional neural networks. We prove that such variance-encoded sparse image datasets enhance computational efficiency, emphasizing the benefits of kernel heterogeneity to leverage naturalistic and variant input structures and possible applications to improve the throughput of neuromorphic hardware.

Recently, there has been an increase in interest in exploring the hypothesis that neural activity conveys information through precise spiking motifs. To investigate this phenomenon, several algorithms have been proposed to detect such motifs in Single Unit Activity recorded from populations of neurons. Based on the inversion of a generative model of raster plot synthesis, we present a novel detection model. This model derives an optimal detection procedure in the form of logistic regression combined with temporal convolution. Its differentiability allows for a supervised learning approach using gradient descent on the binary cross-entropy loss. To assess the model’s ability to detect spiking motifs in synthetic data, numerical evaluations are performed. This analysis emphasizes the benefits of utilizing spiking motifs instead of traditional firing rate-based population codes. Our learning method was able to successfully recover synthetically generated spiking motifs, indicating its potential for further applications. In the future, we aim to extend this method to real neurobiological data, where the ground truth is unknown, to explore and detect spiking motifs in a more natural and biologically relevant context.