The role of population receptive field size and recurrent processing in learning to “de-crowd”

In crowding, the perception of a target is impeded by surrounding clutter. While traditional models are feedforward and local, there is increasing behavioral and neural evidence for a critical role of recurrent processing across the visual hierarchy in crowding. Our recent fMRI findings suggest that higher visual areas, sensitive to global context, determine whether target and flanker features are combined within a single population receptive field (pRF) or rather the target is separated from the flankers in a smaller pRF, likely through recurrent processing. Here, we investigated the neural correlates of perceptual learning to “de-crowd”. Twelve participants trained on an orientation discrimination task under crowding, with the stimulus always presented in the top-right quadrant. Performance on the crowding condition improved substantially in the trained quadrant, but only to a lesser extent in an untrained quadrant. No significant improvements were seen in a no-crowding condition. We used fMRI to estimate pRF sizes in visual areas V1 to V4, separately in areas representing the four visual quadrants. We used dynamic causal modeling to compare bottom-up, top-down, and recurrent models of learning-related modulation of connectivity between visual areas. We found a substantial decrease in pRF size only in left ventral V3, corresponding to the trained visual quadrant. Learning to de-crowd modulated recurrent connectivity across the visual hierarchy, whereas only bottom-up connectivity was modulated in no-crowding. Our findings suggest that improvements in target discrimination under crowding are mediated by top-down processing, which determines the segmentation of the target and flankers through pRF size adjustment. [We would like to thank our funding sources: Swiss National Science Foundation (NCCR Synapsy, project grant numbers 32003B_135679, 32003B_159780, 324730_192755, CRSK-3_190185, 176153), the Leenaards Foundation, Fondation ROGER DE SPOELBERCH, and the Partridge Foundation.]

Enactive robot vision

Mototaka Suzuki

The complexity of today's autonomous robots poses a major challenge for Artificial Intelligence. These robots are equipped with sophisticated sensors and mechanical abilities that allow them to enter our homes and interact with humans. For example, today's robots are almost all equipped with vision and several of them can move over rough terrain with wheels or legs. The methods developed so far in Artificial Intelligence, however, are not yet ready to cope with the complexity of the information gathered through the robot sensors and the need for rapid action in partially unknown and dynamic environments. In this thesis, I will argue that the apparent complexity of the environment and of the robot brain can be significantly simplified if perception, behavior, and learning are allowed to co-develop on the same time scale. In doing so, robots become sensitive to, and actively exploit, characteristics of the environment that they can tackle within their own computational and physical constraints. This line of work is grounded on philosophical and psychological research showing that perception is an active process mediated by behavior. However, computational models of active vision are very rare and often rely on architectures that are preprogrammed to detect certain characteristics of the environment. Previous work have shown that complex visual tasks, such as position and size invariant shape recognition as well as navigation, can be tackled with remarkably simple neural architectures generated by a coevolutionary process of active vision and feature selection. Behavioral machines equipped with primitive vision systems and direct pathways between visual and motor neurons were evolved while freely interacting with their environments. I proceed further on this line of investigation and describe the application of this methodology in three situations, namely car driving with an omnidirectional camera, goal-oriented navigation of a humanoid robot, and cooperative tasks by two agents. I will show that these systems develop sensitivity to a number of oriented, retinotopic, visual features – oriented edges, corners, height – and a behavioral repertoire to locate, bring, and keep these features in sensitive regions of the vision system that allow them to accomplish their goals. In a second set of experiments, I will show that active vision can be exploited by the robot to perform anticipatory exploration of the environment in a task that requires landmark-based navigation. Evolved robots exploit an internal expectation system that makes use of active exploration to check for expected events in the environment. I will describe a third set of experiments where, in addition to an evolutionary process, the visual system of the robot can develop receptive fields by means of unsupervised Hebbian learning and show that these receptive fields are significantly affected by the behavior of the system and differ from those predicted by most computational models of visual cortex. Finally, I will show that these robots replicate the performance deficiencies observed in experiments of motor deprivation with kitten when they are exposed to the same type of motor deprivations. Furthermore, the analyses of our robot brains suggest an explanation for the deficiencies observed in kitten that have not yet been fully understood.

EPFL2007

(Un)crowding modulates recurrent connectivity in the visual cortex

Vitaly Chicherov, Michael Herzog, Maya Anna Jastrzebowska

Introduction: In crowding, neighboring elements impede the perception of a target. Surprisingly, increasing the number of neighboring elements can decrease crowding, i.e., lead to uncrowding (Manassi et al., 2015). Few neuroimaging studies have explored the neural correlates of crowding (Bi et al., 2009; Anderson et al., 2012; Millin et al., 2013; Chen et al., 2014; Chicherov et al., 2014; Ronconi et al., 2016). The resulting evidence about the neural locus of crowding has been equivocal, but crowding is consistently found to attenuate the BOLD response. Here, we used fMRI to investigate the effects of uncrowding on the BOLD response and effective connectivity in the visual cortex. We expected that the percent BOLD signal change (PSC) would best reflect behavioral differences related to (un)crowding in the brain areas driving the crowding effect. We used dynamic causal modelling (DCM; Friston, 2003) to investigate whether (un)crowding is rather a top-down, bottom-up or recurrent processing phenomenon. In accordance with the notion that crowding depends on grouping, we hypothesized that PSC in higher visual areas would better reflect (un)crowding and that effective connectivity would reflect the involvement of top-down processing in uncrowding. Methods: The fMRI experiment consisted of seven conditions: (1) target only (eight circular oblique target gratings surrounding a central fixation dot, tilted either clockwise (CW) or counterclockwise (CCW), (2) 2-flanker (each of the 8 targets was flanked by an inside and outside vertical grating), (3) 4-flanker (one inside and three outside flanker gratings), (4) annulus-flanker (inside and outside flankers connected into annuli), and (5-7) control conditions corresponding to conditions (2-4) with targets removed. Ten participants were asked to indicate whether target gratings were tilted CW or CCW by pushing one of two buttons or to push a button randomly if targets were not present. We calculated the PSC for each participant, each target region of interest (tg-ROI; target-activated areas in V1–V4 and LOC) and each of the conditions of interest and then correlated PSC with response accuracy. We estimated the effective connectivity between tg-ROIs using bilinear DCM and constructed a group-level parametric empirical Bayesian (PEB) model (Friston et al., 2016) over all the participants. After conducting a search over nested models and computing a Bayesian model average (BMA), we compared input-modulatory connectivity (DCM.B matrix) between crowding conditions in terms of excitatory and inhibitory bottom-up and top-down connectivity. Results: Target discrimination (accuracy) was highest in the single target condition, followed by annulus-flanker, 4-flanker and 2-flanker, respectively. A linear mixed model of PSC revealed significant main effects of condition (F(3,171) = 100.06, p

2019

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