Neurons with two sites of synaptic integration learn invariant representations

Learning sparse and meaningful representations through embodiment

How do humans acquire a meaningful understanding of the world with little to no supervision or semantic labels provided by the environment? Here we investigate embodiment with a closed loop between action and perception as one key component in this process. We take a close look at the representations learned by a deep reinforcement learning agent that is trained with high-dimensional visual observations collected in a 3D environment with very sparse rewards. We show that this agent learns stable representations of meaningful concepts such as doors without receiving any semantic labels. Our results show that the agent learns to represent the action relevant information, extracted from a simulated camera stream, in a wide variety of sparse activation patterns. The quality of the representations learned shows the strength of embodied learning and its advantages over fully supervised approaches.

Is realistic neuronal modeling realistic?

Scientific models are abstractions that aim to explain natural phenomena. A successful model shows how a complex phenomenon arises from relatively simple principles while preserving major physical or biological rules and predicting novel experiments. A model should not be a facsimile of reality; it is an aid for understanding it. Contrary to this basic premise, with the 21st century has come a surge in computational efforts to model biological processes in great detail. Here we discuss the oxymoronic, realistic modeling of single neurons. This rapidly advancing field is driven by the discovery that some neurons don't merely sum their inputs and fire if the sum exceeds some threshold. Thus researchers have asked what are the computational abilities of single neurons and attempted to give answers using realistic models. We briefly review the state of the art of compartmental modeling highlighting recent progress and intrinsic flaws. We then attempt to address two fundamental questions. Practically, can we realistically model single neurons? Philosophically, should we realistically model single neurons? We use layer 5 neocortical pyramidal neurons as a test case to examine these issues. We subject three publically available models of layer 5 pyramidal neurons to three simple computational challenges. Based on their performance and a partial survey of published models, we conclude that current compartmental models are ad hoc, unrealistic models functioning poorly once they are stretched beyond the specific problems for which they were designed. We then attempt to plot possible paths for generating realistic single neuron models.

Inter-synaptic learning of combination rules in a cortical network model

Selecting responses in working memory while processing combinations of stimuli depends strongly on their relations stored in long-term memory. However, the learning of XOR-like combinations of stimuli and responses according to complex rules raises the issue of the non-linear separability of the responses within the space of stimuli. One proposed solution is to add neurons that perform a stage of non-linear processing between the stimuli and responses, at the cost of increasing the network size. Based on the non-linear integration of synaptic inputs within dendritic compartments, we propose here an inter-synaptic (IS) learning algorithm that determines the probability of potentiating/depressing each synapse as a function of the co-activity of the other synapses within the same dendrite. The IS learning is effective with random connectivity and without either a priori wiring or additional neurons. Our results show that IS learning generates efficacy values that are sufficient for the processing of XOR-like combinations, on the basis of the sole correlational structure of the stimuli and responses. We analyze the types of dendrites involved in terms of the number of synapses from pre-synaptic neurons coding for the stimuli and responses. The synaptic efficacy values obtained show that different dendrites specialize in the detection of different combinations of stimuli. The resulting behavior of the cortical network model is analyzed as a function of inter-synaptic vs. Hebbian learning. Combinatorial priming effects show that the retrospective activity of neurons coding for the stimuli trigger XOR-like combination-selective prospective activity of neurons coding for the expected response. The synergistic effects of inter-synaptic learning and of mixed-coding neurons are simulated. The results show that, although each mechanism is sufficient by itself, their combined effects improve the performance of the network.

Generalization of learning by synchronous waves: from perceptual organization to invariant organization

From a few presentations of an object, perceptual systems are able to extract invariant properties such that novel presentations are immediately recognized. This may be enabled by inferring the set of all representations equivalent under certain transformations. We implemented this principle in a neurodynamic model that stores activity patterns representing transformed versions of the same object in a distributed fashion within maps, such that translation across the map corresponds to the relevant transformation. When a pattern on the map is activated, this causes activity to spread out as a wave across the map, activating all the transformed versions represented. Computational studies illustrate the efficacy of the proposed mechanism. The model rapidly learns and successfully recognizes rotated and scaled versions of a visual representation from a few prior presentations. For topographical maps such as primary visual cortex, the mechanism simultaneously represents identity and variation of visual percepts whose features change through time.

Cortical region interactions and the functional role of apical dendrites

The basal and distal apical dendrites of pyramidal cells occupy distinct cortical layers and are targeted by axons originating in different cortical regions. Hence, apical and basal dendrites receive information from distinct sources. Physiological evidence suggests that this anatomically observed segregation of input sources may have functional significance. This possibility has been explored in various connectionist models that employ neurons with functionally distinct apical and basal compartments. A neuron in which separate sets of inputs can be integrated independently has the potential to operate in a variety of ways not possible for the conventional neuron model, in which all inputs are treated equally. This article thus considers how functionally distinct apical and basal dendrites can contribute to the information-processing capacities of single neurons and, in particular, how information from different cortical regions could have disparate effects on neural activity and learning.

Slowness: an objective for spike-timing-dependent plasticity?

Our nervous system can efficiently recognize objects in spite of changes in contextual variables such as perspective or lighting conditions. Several lines of research have proposed that this ability for invariant recognition is learned by exploiting the fact that object identities typically vary more slowly in time than contextual variables or noise. Here, we study the question of how this "temporal stability" or "slowness" approach can be implemented within the limits of biologically realistic spike-based learning rules. We first show that slow feature analysis, an algorithm that is based on slowness, can be implemented in linear continuous model neurons by means of a modified Hebbian learning rule. This approach provides a link to the trace rule, which is another implementation of slowness learning. Then, we show analytically that for linear Poisson neurons, slowness learning can be implemented by spike-timing-dependent plasticity (STDP) with a specific learning window. By studying the learning dynamics of STDP, we show that for functional interpretations of STDP, it is not the learning window alone that is relevant but rather the convolution of the learning window with the postsynaptic potential. We then derive STDP learning windows that implement slow feature analysis and the "trace rule." The resulting learning windows are compatible with physiological data both in shape and timescale. Moreover, our analysis shows that the learning window can be split into two functionally different components that are sensitive to reversible and irreversible aspects of the input statistics, respectively. The theory indicates that irreversible input statistics are not in favor of stable weight distributions but may generate oscillatory weight dynamics. Our analysis offers a novel interpretation for the functional role of STDP in physiological neurons.

Learning of somatosensory representations for texture discrimination using a temporal coherence principle

In order to perform appropriate actions, animals need to quickly and reliably classify their sensory input. How can representations suitable for classification be acquired from statistical properties of the animal's natural environment? Akin to behavioural studies in rats, we investigate this question using texture discrimination by the vibrissae system as a model. To account for the rat's active sensing behaviour, we record whisker movements in a hardware model. Based on these signals, we determine the response of primary neurons, modelled as spatio-temporal filters. Using their output, we train a second layer of neurons to optimise a temporal coherence objective function. The performance in classifying textures using a single cell strongly correlates with the cell's temporal coherence; hence output cells outperform primary cells. Using a simple, unsupervised classifier, the performance on the output cell population is same as if using a sophisticated supervised classifier on the primary cells. Our results demonstrate that the optimisation of temporal coherence yields a representation that facilitates subsequent classification by selectively conveying relevant information.

Learning viewpoint invariant perceptual representations from cluttered images

In order to perform object recognition, it is necessary to form perceptual representations that are sufficiently specific to distinguish between objects, but that are also sufficiently flexible to generalize across changes in location, rotation, and scale. A standard method for learning perceptual representations that are invariant to viewpoint is to form temporal associations across image sequences showing object transformations. However, this method requires that individual stimuli be presented in isolation and is therefore unlikely to succeed in real-world applications where multiple objects can co-occur in the visual input. This paper proposes a simple modification to the learning method that can overcome this limitation and results in more robust learning of invariant representations.

A Temporal Stability Approach to Position and Attention-Shift-Invariant Recognition

Incorporation of visual-related self-action signals can help neural networks learn invariance. We describe a method that can produce a network with invariance to changes in visual input caused by eye movements and covert attention shifts. Training of the network is controlled by signals associated with eye movements and covert attention shifting. A temporal perceptual stability constraint is used to drive the output of the network toward remaining constant across temporal sequences of saccadicmotions and covert attention shifts. We use a four-layer neural network model to perform the position-invariant extraction of local features and temporal integration of invariant presentations of local features in a bottom-up structure. We present results on both simulated data and real images to demonstrate that our network can acquire both position and attention shift invariance.

Dendrites: bug or feature?

The integrative properties of dendrites are determined by a complex mixture of factors, including their morphology, the spatio-temporal patterning of synaptic inputs, the balance of excitation and inhibition, and neuromodulatory influences, all of which interact with the many voltage-gated conductances present in the dendritic membrane. Recent efforts to grapple with this complexity have focused on identifying functional compartments in the dendritic tree, the number and size of which depend on the aspect of dendritic function being considered. We discuss how dendritic compartments and the interactions between them help to enhance the computational power of the neuron and define the rules for the induction of synaptic plasticity.