Biology subjects

Maass, W.

Publications and source records attributed to Maass, W..

24 records · Page 2Linked to original sources

bioRxiv2020

Spike-frequency adaptation provides a long short-term memory to networks of spiking neurons

For solving tasks such as recognizing a song, answering a question, or inverting a sequence of symbols, cortical microcircuits need to integrate and manipulate information that was dispersed over time during the preceding seconds. Creating biologically realistic models for the underlying computations, especially with spiking neurons and for behaviorally relevant integration time spans, is notoriously difficult. We examine the role of spike frequency adaptation in such computations and find that it has a surprisingly large impact. The inclusion of this well known property of a substantial fraction of neurons in the neocortex -- especially in higher areas of the human neocortex -- moves the performance of spiking neural network models for computations on network inputs that are temporally dispersed from a fairly low level up to the performance level of the human brain.

neuroscience↗

bioRxiv2019

Brain computation by assemblies of neurons

Assemblies are large populations of neurons believed to imprint memories, concepts, words and other cognitive information. We identify a repertoire of operations on assemblies. These operations correspond to properties of assemblies observed in experiments, and can be shown, analytically and through simulations, to be realizable by generic, randomly connected populations of neurons with Hebbian plasticity and inhibition. Operations on assemblies include: projection (duplicating an assembly by creating a new assembly in a downstream brain area); reciprocal projection (a variant of projection also entailing synaptic connectivity from the newly created assembly to the original one); association (increasing the overlap of two assemblies in the same brain area to reflect cooccurrence or similarity of the corresponding concepts); merge (creating a new assembly with ample synaptic connectivity to and from two existing ones); and pattern-completion (firing of an assembly, with some probability, in response to the firing of some but not all of its neurons). Our analytical results establishing the plausibility of these operations are proved in a simplified mathematical model of cortex: a finite set of brain areas each containing n excitatory neurons, with random connectivity that is both recurrent (within an area) and afferent (between areas). Within one area and at any time, only k of the n neurons fire -- an assumption that models inhibition and serves to define both assemblies and areas -- while synaptic weights are modified by Hebbian plasticity, as well as homeostasis. Importantly, all neural apparatus needed for the functionality of the assembly operations is created on the fly through the randomness of the synaptic network, the selection of the k neurons with the highest synaptic input, and Hebbian plasticity, without any special neural circuits assumed to be in place. Assemblies and their operations constitute a computational model of the brain which we call the Assembly Calculus, occupying a level of detail intermediate between the level of spiking neurons and synapses, and that of the whole brain. As with high-level programming languages, a computation in the Assembly Calculus (that is, a coherent sequence of assembly operations accomplishing a task) can ultimately be reduced -- "compiled down" -- to computation by neurons and synapses; however, it would be far more cumbersome and opaque to represent the same computation that way. The resulting computational system can be shown, under assumptions, to be in principle capable of carrying out arbitrary computations. We hypothesize that something like it may underlie higher human cognitive functions such as reasoning, planning, and language. In particular, we propose a plausible brain architecture based on assemblies for implementing the syntactic processing of language in cortex, which is consistent with recent experimental results.

neuroscience↗

bioRxiv2019

A solution to the learning dilemma for recurrent networks of spiking neurons

Recurrently connected networks of spiking neurons underlie the astounding information processing capabilities of the brain. But in spite of extensive research, it has remained open how they can learn through synaptic plasticity to carry out complex network computations. We argue that two pieces of this puzzle were provided by experimental data from neuroscience. A new mathematical insight tells us how these pieces need to be combined to enable biologically plausible online network learning through gradient descent, in particular deep reinforcement learning. This new learning method - called e-prop - approaches the performance of BPTT (backpropagation through time), the best known method for training recurrent neural networks in machine learning. In addition, it suggests a method for powerful on-chip learning in novel energy-efficient spike-based hardware for AI.

neuroscience↗

bioRxiv2017

Inhibitory networks orchestrate the self-organization of computational function in cortical microcircuit motifs through STDP

Interneurons have diverse morphological and physiological characteristics that potentially contribute to the emergence of powerful computational properties of cortical networks. We investigate the functional role of inhibitory subnetworks in the arguably most common network motif of cortical microcircuits: ensembles of pyramidal cells (PCs) with lateral inhibition, commonly referred to as Winner-Take-All networks. Recent theoretical work has shown that spike-timing-dependent plasticity installs in this network motif an important and ubiquitously useful self-organization process: The emergence of sparse codes and Bayesian inference for repeatedly occurring high-dimensional input patterns. However, this link has so far only been established for strongly simplified models with a symbolic implementation of lateral inhibition, rather than through the interaction of PCs with known types of interneurons. We close this gap in this article, and show that the interaction of PCs with two types of inhibitory networks, that reflect salient properties of somatic-targeting neurons (e.g. basket cells) and dendritic-targeting neurons (e.g. Martinotti cells), provides a good approximation to the theoretically optimal lateral inhibition needed for the self-organization of these network motifs. We provide a step towards unraveling the functional roles of interacting networks of excitatory and inhibitory neurons from the perspective of emergent neural computation.

neuroscience↗

bioRxiv2017

Associations between memory traces emerge in a generic neural circuit model through STDP

Memory traces and associations between them are fundamental for cognitive brain function. Neuron recordings suggest that distributed assemblies of neurons in the brain serve as memory traces for spatial information, real-world items, and concepts. How-ever, there is conflicting evidence regarding neural codes for associated memory traces. Some studies suggest the emergence of overlaps between assemblies during an association, while others suggest that the assemblies themselves remain largely unchanged and new assemblies emerge as neural codes for associated memory items. Here we study the emergence of neural codes for associated memory items in a generic computational model of recurrent networks of spiking neurons with a data-constrained rule for spike-timing-dependent plasticity (STDP). The model depends critically on two parameters, which control the excitability of neurons and the scale of initial synaptic weights. By modifying these two parameters, the model can reproduce both experimental data from the human brain on the fast formation of associations through emergent overlaps between assemblies, and rodent data where new neurons are recruited to encode the associated memories. Hence our findings suggest that the brain can use both of these two neural codes for associations, and dynamically switch between them during consolidation.

neuroscience↗

bioRxiv2016

Searching for Principles of Brain Computation

Highlights O_LIHints for computational principles from experimental data\nC_LIO_LIComputational role of diverse network components\nC_LIO_LIEmergence and computational role of assemblies\nC_LIO_LIProbabilistic inference through stochastic network dynamics\nC_LIO_LIOngoing network rewiring and compensation through synaptic sampling\nC_LI\n\nAbstractExperimental methods in neuroscience, such as calcium-imaging and recordings with multielectrode arrays, are advancing at a rapid pace. They produce insight into the simultaneous activity of large numbers of neurons, and into plasticity processes in the brains of awake and behaving animals. These new data constrain models for neural computation and network plasticity that underlie perception, cognition, behavior, and learning. I will discuss in this short article four such constraints: Inherent recurrent network activity and heterogeneous dynamic properties of neurons and synapses, stereotypical spatio-temporal activity patterns in networks of neurons, high trial-to-trial variability of network responses, and functional stability in spite of permanently ongoing changes in the network. I am proposing that these constraints provide hints to underlying principles of brain computation and learning.

neuroscience↗