A large number of human psychophysical results have been successfully explained in recent years using Bayesian models. However, the neural implementation of such mod-els remains largely unclear. In this paper, we show that a network architecture com-monly used to model the cerebral cortex can implement Bayesian inference for an arbi-trary hidden Markov model. We illustrate the approach using an orientation discrimi-nation task and a visual motion detection task. In the case of orientation discrimination, we show that the model network can infer the posterior distribution over orientations and correctly estimate stimulus orientation in the presence of significant noise. In the case of motion detection, we show that the resulting model network exhibits direction selectivity and correctly computes the posterior probabilities over motion direction and position. When used to solve the well-known random dots motion discrimination task, the model generates responses that mimic the activities of evidence-accumulating neu-rons in cortical areas LIP and FEF. The framework introduced in the paper posits a new interpretation of cortical activities in terms of log posterior probabilities of stimuli occurring in the natural world. 1 1

Soft constraints hypothesis (SCH) is a rational analysis approach that holds that the mixture of perceptual-motor and cognitive resources allocated for interactive behavior is adjusted based on temporal cost-benefit tradeoffs. Alternative approaches maintain that cognitive resources are in some sense protected or conserved in that greater amounts of perceptual-motor effort will be expended to conserve lesser amounts of cognitive effort. One alternative, the minimum memory hypothesis (MMH), holds that people favor strategies that minimize the use of memory. SCH is compared with MMH across 3 experiments and with predictions of an Ideal Performer Model that uses ACT-R’s memory system in a reinforcement learning approach that maximizes expected utility by minimizing time. Model and data support the SCH view of resource allocation; at the under 1000-ms level of analysis, mixtures of cognitive and perceptual-motor resources are adjusted based on their cost-benefit tradeoffs for interactive behavior.

In Pavlovian and instrumental conditioning, reward typically comes seconds after reward-triggering actions, creating an explanatory conundrum known as ‘‘distal reward problem’’: How does the brain know what firing patterns of what neurons are responsible for the reward if 1) the patterns are no longer there when the reward arrives and 2) all neurons and synapses are active during the waiting period to the reward? Here, we show how the conundrum is resolved by a model network of cortical spiking neurons with spike-timing--dependent plasticity (STDP) modulated by dopamine (DA). Although STDP is triggered by nearly coincident firing patterns on a millisecond timescale, slow kinetics of subsequent synaptic plasticity is sensitive to changes in the extracellular DA concentration during the critical period of a few seconds. Random firings during the waiting period to the reward do not affect STDP and hence make the network insensitive to the ongoing activity— the key feature that distinguishes our approach from previous theoretical studies, which implicitly assume that the network be quiet during the waiting period or that the patterns be preserved until the reward arrives. This study emphasizes the importance of precise firing patterns in brain dynamics and suggests how a global diffusive reinforcement signal in the form of extracellular DA can selectively influence the right synapses at the right time.

This article is about how the brain data mines its sensory inputs. There are several architectural principles of functional brain anatomy that have emerged from careful anatomic and physiologic studies over the past century. These principles are considered in the light of representational learning to see if they could have been predicted a priori on the basis of purely theoretical considerations. We first review the organisation of hierarchical sensory cortices, paying special attention to the distinction between forward and backward connections. We then review various approaches to representational learning as special cases of generative models, starting with supervised learning and ending with learning based upon empirical Bayes. The latter predicts many features, such as a hierarchical cortical system, prevalent top-down backward influences and functional asymmetries between forward and backward connections that are seen in the real brain. The key points made in this article are: (i) hierarchical generative models enable the learning of empirical priors and eschew prior assumptions about the causes of sensory input that are inherent in non-hierarchical models. These assumptions are necessary for learning schemes based on information theory and efficient or sparse coding, but are not necessary in a hierarchical context. Critically, the anatomical infrastructure that may implement generative models in the brain is hierarchical. Furthermore, learning based on empirical Bayes can proceed in a biologically plausible way. (ii) The second point is that backward connections are essential if the processes generating inputs cannot be inverted, or the inversion cannot be parameterised. Because these processes involve many-to-one mappings, are non-linear and dynamic in nature, they are generally non-invertible. This enforces an explicit parameterisation of generative models (i.e. backward

Abstract not found

Development imposes great challenges. Internal “cortical” representations must be autonomously generated from interactive experiences. The eventual quality of these developed representations is of course important. Additionally, learning must be as fast as possible—to quickly derive better representation from limited experiences. Those who achieve both of these will have competitive advantages. We present a cortex-inspired theory called lobe component analysis (LCA) guided by the aforementioned dual criteria. A lobe component represents a high concentration of probability density of the neuronal input space. We explain how lobe components can achieve a dual—spatiotemporal (“best ” and “fastest”)—optimality, through mathematical analysis, in which we describe how lobe components ’ plasticity can be temporally scheduled to take into account the history of observations in the best possible way. This contrasts with using only the last observation in gradient-based adaptive learning algorithms. Since they are based on two cell-centered mechanisms—Hebbian learning and lateral inhibition—lobe components develop in-place, meaning every networked neuron is individually responsible for the learning of its signal-processing characteristics within its connected network environment. There is no need for a separate learning network. We argue that in-place learning algorithms will be crucial for real-world large-size developmental applications due to their simplicity, low computational complexity, and generality. Our experimental results show that the learning speed of the LCA algorithm is drastically faster than other Hebbian-based updating methods and independent component analysis algorithms, thanks to its dual optimality, and it does not need to use any second- or higher order statistics. We also introduce the new principle of fast learning from stable representation.

The time histogram method is the most basic tool for capturing a timedependent rate of neuronal spikes. Generally in the neurophysiological literature, the bin size that critically determines the goodness of the fit of the time histogram to the underlying spike rate has been subjectively selected by individual researchers. Here, we propose a method for objectively selecting the bin size from the spike count statistics alone, so that the resulting bar or line graph time histogram best represents the unknown underlying spike rate. For a small number of spike sequences generated from a modestly fluctuating rate, the optimal bin size may diverge, indicating that any time histogram is likely to capture a spurious rate. Given a paucity of data, the method presented here can nevertheless suggest how many experimental trials should be added in order to obtain a meaningful time-dependent histogram with the required accuracy.

We present a mathematical analysis of a networks with Integrate-and-Fire neurons with conductance based synapses. Taking into account the realistic fact that the spike time is only known within some finite precision, we propose a model where spikes are effective at times multiple of a characteristic time scale δ, where δ can be arbitrary small (in particular, well beyond the numerical precision). We make a complete mathematical characterization of the model-dynamics and obtain the following results. The asymptotic dynamics is composed by finitely many stable periodic orbits, whose number and period can be arbitrary large and can diverge in a region of the synaptic weights space, traditionally called the “edge of chaos”, a notion mathematically well defined in the present paper. Furthermore, except at the edge of chaos, there is a one-to-one correspondence between the membrane potential trajectories and the raster plot. This shows that the neural code is entirely “in the spikes ” in this case. As a key tool, we introduce an order parameter, easy to compute numerically, and closely related to a natural notion of entropy, providing a relevant characterization of the computational capabilities of the network. This allows us to compare the computational capabilities of leaky and Integrate-and-Fire models and conductance based models. The present study considers networks with constant input, and without time-dependent plasticity, but the framework has been designed for both extensions.

Abstract: A wealth of research focuses on the decision-making processes that animals and humans employ when selecting actions in the face of reward and punishment. Initially such work stemmed from psychological investigations of conditioned behavior, and explanations of these in terms of computational models. Increasingly, analysis at the computational level has drawn on ideas from reinforcement learning, which provide a normative framework within which decision-making can be analyzed. More recently, the fruits of these extensive lines of research have made contact with investigations into the neural basis of decision making. Converging evidence now links reinforcement learning to specific neural substrates, assigning them precise computational roles. Specifically, electrophysiological recordings in behaving animals and functional imaging of human decision-making have revealed in the brain the existence of a key reinforcement learning signal, the temporal difference reward prediction error. Here, we first introduce the formal reinforcement learning framework. We then review the multiple lines of evidence linking reinforcement learning to the function of dopaminergic neurons in the mammalian midbrain and

The role of feedback in perceptual learning is probed in an orientation discrimination experiment under destabilizing non-stationary conditions, and explored in a neural-network model. Experimentally, perceptual learning was examined with periodic alteration of a strong external noise context. The speed of learning, the performance loss at each change in external noise context (switch cost), and the asymptotic accuracy d 0 without feedback were very similar or identical to those with feedback. However, lack of feedback led to higher decision bias (error responses matching the external noise context). In the model, the stimulus representations are constant, whereas the read-out connections to a decision unit learn by a Hebbian plasticity rule that may be augmented by additional feedback input and criterion control of decision bias.