Neural networks

…brain-based treatments that target neural networks involved in cognitive control and emotion regulation.

…design, complex biological traits and neural networks, with dual supervision by fellows from complementary backgrounds. The programme spans more than 30 research groups across Germany and includes annual interdisciplinary schools, conference funding and mentoring. Applications close on

Apply neural networks to biological sequence analysis, including sequence classification, functional-feature discovery, source tracking and sequence generation. Hands-on notebooks introduce sequence encoding and language-model concepts, progressing from convolutional networks and LSTMs to attention and Transformers. Exercises use TensorFlow and Keras to connect model training and performance evaluation with the biological interpretation of genomic data. 2027-01-25–2027-01-27, 14:00–20:00 daily, Europe/Berlin (CET, UTC+01:00). Online course; fee €480. Participants should be familiar with Pytho

…Neural networks can construct crossing-symmetric correlators from limited physical input; their tendency toward simple, smooth solutions may reveal additional organizing principles behind bootstrap constraints.

Tom Griffiths discusses mathematical accounts of cognition through logic, neural networks and probability. The conversation draws on his book The Laws of Thought to examine how these approaches complement one another, the ideas and people behind them, and prospects for a unified theory of mind.

…how both artificial and natural networks of spiking neurons can learn energy-efficient information processing strategies. Here we propose Predictive Coding Light (PCL), a recurrent hierarchical spiking neural network for unsupervised representation learning. In contrast to previous predictive coding approaches, PCL does not transmit prediction errors to higher

Christian Ferko reviews correspondences between neural networks, quantum mechanics and quantum field theory. Wide networks behave like Gaussian processes and share features with free or weakly interacting fields. Focusing on one-dimensional field theories gives enough control to show, under mild assumptions, that any quantum-particle model has a neural-network representation. Architectural choices that satisfy quantum axioms then produce properties such as uncertainty relations. The presentation draws on joint work with Jim Halverson.

…understand how biophysically realistic neural networks support these fundamental brain functions, we developed an analytical framework based on the principle of efficient coding (Barlow,

…With the rise of neural networks (NN) that exhibit a striking apparent proficiency for abstractions, equality reasoning in NNs has also gained interest. Yet despite the extensive study, conclusions about equality reasoning vary widely. Some authors claim that SD can be learned

The critical brain hypothesis proposes that neural networks operate near criticality to reap the computational benefits of accessing a wide range of timescales [1]. Proponents of this viewpoint highlight the presence of heavy-tailed (power-law) spatiotemporal correlations as markers of criticality in the brain [2]. Critics of this hypothesis argue that such correlations could be inherited from upstream sources, such as sensory input [3]. Similarly, Ref. [4] constructed a model of independent neurons driven by shared noise input that exhibited neural activity with power-law tails, which could b

Uncertainty calibration, the process of aligning a model’s predictive confidence with the likelihood of correctness, is crucial for artificial intelligence systems. Despite recent advancements in deep learning, conventional models often struggle to calibrate confidence in a manner similar to human judgment (Guo et al., 2017). Specifically, while increased model capacity facilitates accurate classification, it frequently undermines confidence calibration. For instance, these models tend to exhibit overconfidence when confronted with unknown outlier samples, even in the absence of relevant knowl

…computational approaches, including neural networks, for recognising regulatory DNA and potentially designing synthetic elements. It also compares recurring arrangements of enhancers around genes important in mammalian development.

…For neural networks, this enables privacy-oriented obfuscation and interpretable estimates of topological phases. The method also supplies an efficient initialisation for tensor-train optimisation and improves the memory–computation trade-off in model compression.

Behavior and cognition depend on the integrated action of neural structures and populations distributed throughout the brain. We recently developed a set of molecular imaging tools that enable multiregional processing and plasticity in neural networks to be studied at a brain-wide scale in rodents and nonhuman primates. Here we will describe how a novel genetically encoded activity reporter enables information flow in virally labeled neural circuitry to be monitored by fMRI. Using the reporter to perform functional imaging of synaptically defined neural populations in the rat somatosensory sys

A hallmark of biological and artificial neural networks is that neurons tile the range of continuous sensory inputs and intrinsic variables with overlapping responses. It is characteristic for the underlying recurrent connectivity in the cortex that neurons with similar tuning predominantly excite each other. The reason for such an architecture is not clear. Using an analytically tractable model, we show that it can naturally arise from a cooperative coding scheme. In this scheme neurons with similar responses specifically support each other by sharing their computations to obtain the desired

Automated voltage-clamp recordings from 40 voltage-gated potassium (Kv) channels [1] revealed the inherent kinetic heterogeneity of some Kv types expressed in the brain (e.g. Kv1.3). A model capable of capturing the kinetic Kv heterogeneity requires both fixed model features shared between cells, and the ability to capture the variability present in individual cells. The classical Hodgkin Huxley (HH) formalism [2], by itself, does not possess these properties. We apply the Deep Non-Linear Mixed Effects (DeepNLME) framework [3] to model the kinetic Kv heterogeneity. DeepNLME combines deep lea

The effects of a global inhibition are studied on different models of neural networks, from the Wilson-Cowan model to networks of leaky integrate-and-fire neurons. These systems typically display a region of parameter space where two different states are possible, one of low and another of high activity with a phenomenology typical of first order phase transitions, with e.g. hysteresis in correspondence of a sweep of a control parameter. The introduction of a global inhibition, given by the presence of inhibitory interneurons with very large in-degree and out-degree, leads to a smoothing of t

Understanding the fine-scale connectivity structure of neural networks is key to deciphering their behaviour. Motifs in network connectivity are instrumental in relating circuit structure to dynamical properties of neural activity, such as covariability and dimensionality [1]. Experiments have made significant progress in measuring precise neural connectivity: multi-patch experiments, as well as more recent studies in connectomics, have found that certain types of three-neuron (triplet) connectivity motifs are over- or underrepresented relative to chance (Fig.1A, [2,3]). However, it is still u

Animals continuously merge information across their senses, and use these combined signals to guide their behaviour. In the brain, these multisensory computations are performed by circuits rich with complex connections like skips [1] - bypassing certain layers, and feedback pathways which allow information to flow from later (multisensory) to earlier (‘unisensory’) parts of the network [2]. However, pure-feedforward models have been shown to be optimal for many multisensory tasks [3,4], commonly used to test experimental subjects or computational models; raising the question, what role do non-

…Software Module which uses trained neural networks enabling label-free analysis of phase contrast images through accurate segmentation of individual cells and Live/Dead cl

…and dynamics within neural networks, which underpin cognitivefunctions and behaviors in biological organisms. The Blue Brain Cellular Lab-oratory (BlueCelluLab) is a software package designed for detailed simulationand experimentation on individual and groups of neurons.Key features of BlueCelluLab include the ability to automate in-silico whole-cell patching experiments, offering a method for exploring neuron behavior in acontrolled, virtual setting. The software is effective in facilitating the validationo

The patch-clamp technique allows us to eavesdrop the gating behavior of individual ion channels with unprecedented temporal resolution. The recorded signals arise from conformational changes of the protein or protein complex forming the ion channels as they make rapid transitions between conducting and non-conducting states. However, unambiguous analysis of single-channel datasets is challenging given the inadvertently low signal-to-noise ratio as well as signal distortions caused by low-pass filtering and limited recording bandwidth. Ion channel kinetics are typically described using hidden-M

Dendritic spines are the seat of most excitatory synapses in the brain, and a cellular structure considered central to learning, memory, and activity-dependent plasticity. Typically, humans quantify dendritic spines from light microscopy data in a manual painstaking, error-prone process. We found that human-to-human variability is substantial (inter-rater reliability 82.2±6.4 %), raising concerns about the reproducibility of experiments and the validity of using human-annotated 'ground truth' as an evaluation method for computational approaches of spine identification. To address this, we pres

Elements of neural networks (NNs), both biological and artificial, can be described by selectivity to specific cognitive features. Revealing such features is important for understanding NNs' inner working mechanisms. For living systems, where the neural response to a stimulus is not a known function, one needs to build a feedback loop of exposing it to a set of different stimuli. The properties of such stimuli should be iteratively varied aiming in the direction of neuronal maximal response. To utilize such a loop for a biological network, one should run it quickly and efficiently, reaching th