Models of Evidence Integration in Rapid Decision Making Processes

Decision making is of crucial interest in many disciplines such as psychology, neuroscience, economics and machine learning. Binary perceptual decision theories relate to situations where an observer (or machine) is confronted with one of two possible noisy stimuli. For example, human readers have to decide whether a handwritten character is an n or a u; a trader has to decide whether to sell or to keep; a monkey has to decide whether dots on a screen are moving to the left or to the right. While engineering and economical decision theories focus on how to compute optimal decisions, psychology and neuroscience investigate the actual decision making process in humans and animals. Humans only need a fraction of a second to recognize objects. This astonishing speed is also evident in sports such as table tennis or soccer requiring rapid reactions to moving balls. In these tasks, the brain has to decide rapidly upon visual information available for only a hundred milliseconds or less. However, most experimental and theoretical work on decision making focuses on stationary paradigms and surprisingly few studies have taken the timing of the stimulus into account. Decisions about noisy stimuli require evidence integration over time. Traditionally, evidence integration and decision making are described as a one-stage process: a decision is made when evidence for the presence of a stimulus crosses a threshold. We will show that this model is incompatible with psychophysical experiments on feature fusion, where two visual stimuli are presented in rapid succession. Paradoxically, the second stimulus biases decisions more strongly than the first one, contrary to predictions of one-stage models and intuition. We present a two-stage model where sensory information is integrated and buffered before it starts driving the diffusion process.

Biologically plausible unsupervised learning in shallow and deep neural networks

Bernd Albert Illing

The way our brain learns to disentangle complex signals into unambiguous concepts is fascinating but remains largely unknown. There is evidence, however, that hierarchical neural representations play a key role in the cortex. This thesis investigates biologically plausible models of unsupervised learning of hierarchical representations as found in the brain and modern computer vision models. We use computational modeling to address three main questions at the intersection of artificial intelligence (AI) and computational neuroscience.The first question is: What are useful neural representations and when are deep hierarchical representations needed? We approach this point with a systematic study of biologically plausible unsupervised feature learning in a shallow 2-layer networks on digit (MNIST) and object (CIFAR10) classification. Surprisingly, random features support high performance, especially for large hidden layers. When combined with localized receptive fields, random feature networks approach the performance of supervised backpropagation on MNIST, but not on CIFAR10. We suggest that future models of biologically plausible learning should outperform such random feature benchmarks on MNIST, or that such models should be evaluated in different ways.The second question is: How can hierarchical representations be learned with mechanisms supported by neuroscientific evidence? We cover this question by proposing a unifying Hebbian model, inspired by common models of V1 simple and complex cells based on unsupervised sparse coding and temporal invariance learning. In shallow 2-layer networks, our model reproduces learning of simple and complex cell receptive fields, as found in V1. In deeper networks, we stack multiple layers of Hebbian learning but find that it does not yield hierarchical representations of increasing usefulness. From this, we hypothesise that standard Hebbian rules are too constrained to build increasingly useful representations, as observed in higher areas of the visual cortex or deep artificial neural networks.The third question is: Can AI inspire learning models that build deep representations and are still biologically plausible? We address this question by proposing a learning rule that takes inspiration from neuroscience and recent advances in self-supervised deep learning. The proposed rule is Hebbian, i.e. only depends on pre- and post-synaptic neuronal activity, but includes additional local factors, namely predictive dendritic input and widely broadcasted modulation factors. Algorithmically, this rule applies self-supervised contrastive predictive learning to a causal, biological setting using saccades. We find that networks trained with this generalised Hebbian rule build deep hierarchical representations of images, speech and video.We see our modeling as a potential starting point for both, new hypotheses, that can be tested experimentally, and novel AI models that could benefit from added biological realism.

EPFL2021

Interpretation of neuronal response properties with simplified neuron models

Laurent Badel

Computational neuroscience is a branch of the neurosciences that attempts to elucidate the principles underlying the operation of neurons with the help of mathematical modeling. In contrast with a number of fields pursuing a tightly related goal, such as machine learning or statistical learning, computational neuroscience emphasizes the realistic description of neuronal function and organization, and therefore requires a collaboration with experimental research in order to design new models and test their predictions. With the growing computational power available in modern computers, the simulation of neuronal networks with realistic numbers of neurons and functional organization is becoming routinely accessible. For the accurate design of such network models, the experimental characterization of neuronal response properties is required to probe the structural layout of biological networks at the single cell level. We develop here a novel method, the dynamic I-V method, that allows for the extraction of cellular response properties, and requires only a small amount of experimental data. Using this technique a simplified model of the neuronal response is obtained that is shown to accurately fit the experimental data. We demonstrate the use of the dynamic I-V method on both cortical pyramidal cells and inhibitory interneurons, and give the distributions of cellular parameters for the studied sample of cells. We also give theoretical results on the spike-triggered average, a quantity that is easily measured experimentally, which relate the neuronal response properties to identifiable features of the spike-triggered average, and can potentially be helpful to extend the dynamic I-V method in order to derive more accurate models. Potential applications and extensions of our results are discussed.

EPFL2009

Biologically plausible unsupervised learning in shallow and deep neural networks

Bernd Albert Illing

EPFL2021