A5040065902- Neural dynamics and brain function
- Advanced Memory and Neural Computing
- Neural Networks and Applications
- Neural Networks and Reservoir Computing
- Olfactory and Sensory Function Studies
- Neuroscience and Music Perception
- Neurobiology and Insect Physiology Research
- EEG and Brain-Computer Interfaces
- Memory and Neural Mechanisms
- Ferroelectric and Negative Capacitance Devices
- Neuroscience and Neuropharmacology Research
- Advanced Chemical Sensor Technologies
- Relativity and Gravitational Theory
- Software Engineering Research
- Cell Image Analysis Techniques
- Evolutionary Algorithms and Applications
- Fractal and DNA sequence analysis
- Anomaly Detection Techniques and Applications
- Quantum optics and atomic interactions
- Cosmology and Gravitation Theories
- Photoreceptor and optogenetics research
- Noncommutative and Quantum Gravity Theories
- Reinforcement Learning in Robotics
- Bayesian Modeling and Causal Inference
- Visual perception and processing mechanisms
Centrum Wiskunde & Informatica
2022-2025
University of Sheffield
2020-2023
University of Bonn
2018-2020
École Polytechnique Fédérale de Lausanne
2017-2018
National Centre for Biological Sciences
2010-2015
Tata Institute of Fundamental Research
2007-2010
Task-switching is a fundamental cognitive ability that allows animals to update their knowledge of current rules or contexts. Detecting discrepancies between predicted and observed events essential for this process. However, little known about how the brain computes prediction-errors whether neural prediction-error signals are causally related task-switching behaviours. Here we trained mice use switch, in single trial, responding same stimuli using two distinct rules. Optogenetic silencing...
The brain needs to predict how the body reacts motor commands, but a network of spiking neurons can learn non-linear dynamics using local, online and stable learning rules is unclear. Here, we present supervised scheme for feedforward recurrent connections in heterogeneous neurons. error output fed back through fixed random with negative gain, causing follow desired dynamics. rule Feedback-based Online Local Learning Of Weights (FOLLOW) local sense that weight changes depend on presynaptic...
The ability of humans and animals to quickly adapt novel tasks is difficult reconcile with the standard paradigm learning by slow synaptic weight modification. Here we show that fixed-weight neural networks can learn generate required dynamics imitation. After appropriate pretraining, dynamically new thereafter continue achieve them without further teacher feedback. We explain this illustrate it a variety target dynamics, ranging from oscillatory trajectories driven chaotic dynamical systems.
The field of neuromorphic computing holds great promise in terms advancing efficiency and capabilities by following brain-inspired principles. However, the rich diversity techniques employed research has resulted a lack clear standards for benchmarking, hindering effective evaluation advantages strengths methods compared to traditional deep-learning-based methods. This paper presents collaborative effort, bringing together members from academia industry, define benchmarks computing:...
The processing of sequential and temporal data is essential to computer vision speech recognition, two the most common applications artificial intelligence (AI). Reservoir computing (RC) a branch AI that offers highly efficient framework for inputs at low training cost compared conventional Recurrent Neural Networks (RNNs). However, despite extensive effort, two-terminal memristor-based reservoirs have, until now, been implemented process by reading their conductance states only once, end...
Stimulus encoding by primary sensory brain areas provides a data-rich context for understanding their circuit mechanisms. The vertebrate olfactory bulb is an input area having unusual two-layer dendro-dendritic connections whose roles in odor coding are unclear. To clarify these roles, we built detailed compartmental model of the rat that synthesizes much wider range experimental observations on bulbar physiology and response dynamics than has hitherto been modeled. We predict...
Abstract Task-switching is a fundamental cognitive ability that allows animals to update their knowledge of current rules or contexts. Detecting discrepancies between predicted and observed events essential for this process. However, little known about how the brain computes prediction-errors whether neural prediction-error signals are causally related task-switching behaviours. Here we trained mice use switch, in single trial, responding same stimuli using two distinct rules. Optogenetic...
Abstract Neuronal manifold learning techniques represent high-dimensional neuronal dynamics in low-dimensional embeddings to reveal the intrinsic structure of manifolds. A common goal these is learn that allow a good reconstruction original data. We introduce novel technique, BunDLe-Net, learns Markovian embedding which pre-serves only those aspects are relevant for given behavioural context. In this way, BunDLe-Net eliminates irrelevant decoding behaviour, effectively de-noising data better...
Learning weights in a spiking neural network with hidden neurons, using local, stable and online rules, to control non-linear body dynamics is an open problem. Here, we employ supervised scheme, Feedback-based Online Local Of Weights (FOLLOW), train of heterogeneous neurons layers, two-link arm so as reproduce desired state trajectory. The first learns inverse model the dynamics, i.e. from trajectory input network, it infer continuous-time command that produced Connection are adjusted via...
Learning and memory are intertwined in our brain their relationship is at the core of several recent neural network models. In particular, Attention-Gated MEmory Tagging model (AuGMEnT) a reinforcement learning with an emphasis on biological plausibility dynamics learning. We find that AuGMEnT does not solve some hierarchical tasks, where higher-level stimuli have to be maintained over long time, while lower-level need remembered forgotten shorter timescale. To overcome this limitation, we...
Learning representations of underlying environmental dynamics from partial observations is a critical challenge in machine learning. In the context Partially Observable Markov Decision Processes (POMDPs), state are often inferred history past and actions. We demonstrate that incorporating future information essential to accurately capture causal enhance representations. To address this, we introduce Dynamical Variational Auto-Encoder (DVAE) designed learn Markovian offline trajectories...
In complex natural environments, sensory systems are constantly exposed to a large stream of inputs. Novel or rare stimuli, which often associated with behaviorally important events, typically processed differently than the steady background, has less relevance. Neural signatures such differential processing, commonly referred as novelty detection, have been identified on level EEG recordings mismatch negativity (MMN) and single neurons stimulus-specific adaptation (SSA). Here, we propose...
The linear mathematics of Fourier composition and decomposition monochromatic electromagnetic fields is supposed to have a direct realization in the physical world sense that we assume complete equivalence reality as well effects when an arbitrary field substituted physically with its components, vice versa, same spatial region. In simplest cases, two superposed light at frequencies &ohgr;<sub>1</sub> &ohgr;<sub>2</sub> are be identical their single average frequency, amplitude modulated...
Abstract In complex natural environments, sensory systems are constantly exposed to a large stream of inputs. Novel or rare stimuli, which often associated with behaviorally important events, typically processed differently than the steady background, has less relevance. Neural signatures such differential processing, commonly referred as novelty detection, have been identified on level EEG recordings mismatch negativity and single neurons stimulus-specific adaptation. Here, we propose...
Implementation of accurate neural network models in edge applications such as wearables is limited by the hardware platform due to constraints power/area. We highlight novel concepts reservoir computing that rely on a volatile three terminal solid electrolyte thin film synaptic transistor, whose conductance can be controlled gate and drain voltages enhance richness operate off-state. The proposed approach achieves an accuracy 94% image processing, significantly higher than equivalent based...
Reservoir computing using delay systems took off when it was demonstrated that a single volatile memristor can be effectively utilized to reduce the burden of training recurrent neural networks, without need for any interconnected reservoir nodes. Here we demonstrate maximum impact on learning efficiency $\mathrm{Z}\mathrm{n}\mathrm{O}/\mathrm{T}\mathrm{a}_{2}\mathrm{O}_{5}$ SEFET based is derived from (i) output after every pulse, (ii) device variability, and (iii) scanning input image data...
Extant proofs of the spin-statistics connection (SSC) are kinematical. C S Unnikrishnan has suggested that a dynamical interaction leading to SSC would involve spin and perforce gravity, only known universal force. For scattering two identical particles, he considers [arXiv: gr-qc/0406043] their spins with gravito-magnetic field generated by motion through cosmic matter-energy. There direct particles-exchanged amplitudes accumulate different quantum phases which provide relevant...
Recently, it was proposed that the spin-statistics connection arises due to a quantum dynamical phase involving cosmic gravity (Unnikrishnan C S, gr-qc/0406043). There assumed gravitational accumulated on wavefunctions of each two identical particles undergoing scattering in presence matter with direction their momentum changing by some angle is equivalent forward entire universe rotating oppositely same angle. However, if one considers an electron orbit be analogous acquired particle,...