Generative Models for Active Vision

The role of prediction and visual tracking strategies during manual interception: An exploration of individual differences

The interception (or avoidance) of moving objects is a common component of various daily living tasks; however, it remains unclear whether precise alignment of foveal vision with a target is important for motor performance. Furthermore, there has also been little examination of individual differences in visual tracking strategy and the use of anticipatory gaze adjustments. We examined the importance of in-flight tracking and predictive visual behaviors using a virtual reality environment that required participants (n = 41) to intercept tennis balls projected from one of two possible locations. Here, we explored whether different tracking strategies spontaneously arose during the task, and which were most effective. Although indices of closer in-flight tracking (pursuit gain, tracking coherence, tracking lag, and saccades) were predictive of better interception performance, these relationships were rather weak. Anticipatory gaze shifts toward the correct release location of the ball provided no benefit for subsequent interception. Nonetheless, two interceptive strategies were evident: 1) early anticipation of the ball's onset location followed by attempts to closely track the ball in flight (i.e., predictive strategy); or 2) positioning gaze between possible onset locations and then using peripheral vision to locate the moving ball (i.e., a visual pivot strategy). Despite showing much poorer in-flight foveal tracking of the ball, participants adopting a visual pivot strategy performed slightly better in the task. Overall, these results indicate that precise alignment of the fovea with the target may not be critical for interception tasks, but that observers can adopt quite varied visual guidance approaches.

An Overview of the Free Energy Principle and Related Research

The free energy principle and its corollary, the active inference framework, serve as theoretical foundations in the domain of neuroscience, explaining the genesis of intelligent behavior. This principle states that the processes of perception, learning, and decision making-within an agent-are all driven by the objective of "minimizing free energy," evincing the following behaviors: learning and employing a generative model of the environment to interpret observations, thereby achieving perception, and selecting actions to maintain a stable preferred state and minimize the uncertainty about the environment, thereby achieving decision making. This fundamental principle can be used to explain how the brain processes perceptual information, learns about the environment, and selects actions. Two pivotal tenets are that the agent employs a generative model for perception and planning and that interaction with the world (and other agents) enhances the performance of the generative model and augments perception. With the evolution of control theory and deep learning tools, agents based on the FEP have been instantiated in various ways across different domains, guiding the design of a multitude of generative models and decision-making algorithms. This letter first introduces the basic concepts of the FEP, followed by its historical development and connections with other theories of intelligence, and then delves into the specific application of the FEP to perception and decision making, encompassing both low-dimensional simple situations and high-dimensional complex situations. It compares the FEP with model-based reinforcement learning to show that the FEP provides a better objective function. We illustrate this using numerical studies of Dreamer3 by adding expected information gain into the standard objective function. In a complementary fashion, existing reinforcement learning, and deep learning algorithms can also help implement the FEP-based agents. Finally, we discuss the various capabilities that agents need to possess in complex environments and state that the FEP can aid agents in acquiring these capabilities.

Object-Centric Scene Representations Using Active Inference

Representing a scene and its constituent objects from raw sensory data is a core ability for enabling robots to interact with their environment. In this letter, we propose a novel approach for scene understanding, leveraging an object-centric generative model that enables an agent to infer object category and pose in an allocentric reference frame using active inference, a neuro-inspired framework for action and perception. For evaluating the behavior of an active vision agent, we also propose a new benchmark where, given a target viewpoint of a particular object, the agent needs to find the best matching viewpoint given a workspace with randomly positioned objects in 3D. We demonstrate that our active inference agent is able to balance epistemic foraging and goal-driven behavior, and quantitatively outperforms both supervised and reinforcement learning baselines by more than a factor of two in terms of success rate.

How 'visual' is the visual cortex? The interactions between the visual cortex and other sensory, motivational and motor systems as enabling factors for visual perception

The definition of the visual cortex is primarily based on the evidence that lesions of this area impair visual perception. However, this does not exclude that the visual cortex may process more information than of retinal origin alone, or that other brain structures contribute to vision. Indeed, research across the past decades has shown that non-visual information, such as neural activity related to reward expectation and value, locomotion, working memory and other sensory modalities, can modulate primary visual cortical responses to retinal inputs. Nevertheless, the function of this non-visual information is poorly understood. Here we review recent evidence, coming primarily from studies in rodents, arguing that non-visual and motor effects in visual cortex play a role in visual processing itself, for instance disentangling direct auditory effects on visual cortex from effects of sound-evoked orofacial movement. These findings are placed in a broader framework casting vision in terms of predictive processing under control of frontal, reward- and motor-related systems. In contrast to the prevalent notion that vision is exclusively constructed by the visual cortical system, we propose that visual percepts are generated by a larger network-the extended visual system-spanning other sensory cortices, supramodal areas and frontal systems. This article is part of the theme issue 'Decision and control processes in multisensory perception'.

An investigation of feed-forward and feedback eye movement training in immersive virtual reality

The control of eye gaze is critical to the execution of many skills. The observation that task experts in many domains exhibit more efficient control of eye gaze than novices has led to the development of gaze training interventions that teach these behaviours. We aimed to extend this literature by i) examining the relative benefits of feed-forward (observing an expert's eye movements) versus feed-back (observing your own eye movements) training, and ii) automating this training within virtual reality. Serving personnel from the British Army and Royal Navy were randomised to either feed-forward or feed-back training within a virtual reality simulation of a room search and clearance task. Eye movement metrics - including visual search, saccade direction, and entropy - were recorded to quantify the efficiency of visual search behaviours. Feed-forward and feed-back eye movement training produced distinct learning benefits, but both accelerated the development of efficient gaze behaviours. However, we found no evidence that these more efficient search behaviours transferred to better decision making in the room clearance task. Our results suggest integrating eye movement training principles within virtual reality training simulations may be effective, but further work is needed to understand the learning mechanisms.

Symmetry and complexity in object-centric deep active inference models

Humans perceive and interact with hundreds of objects every day. In doing so, they need to employ mental models of these objects and often exploit symmetries in the object's shape and appearance in order to learn generalizable and transferable skills. Active inference is a first principles approach to understanding and modelling sentient agents. It states that agents entertain a generative model of their environment, and learn and act by minimizing an upper bound on their surprisal, i.e. their free energy. The free energy decomposes into an accuracy and complexity term, meaning that agents favour the least complex model that can accurately explain their sensory observations. In this paper, we investigate how inherent symmetries of particular objects also emerge as symmetries in the latent state space of the generative model learnt under deep active inference. In particular, we focus on object-centric representations, which are trained from pixels to predict novel object views as the agent moves its viewpoint. First, we investigate the relation between model complexity and symmetry exploitation in the state space. Second, we do a principal component analysis to demonstrate how the model encodes the principal axis of symmetry of the object in the latent space. Finally, we also demonstrate how more symmetrical representations can be exploited for better generalization in the context of manipulation.

A Bayesian computational model to investigate expert anticipation of a seemingly unpredictable ball bounce

During dynamic and time-constrained sporting tasks performers rely on both online perceptual information and prior contextual knowledge to make effective anticipatory judgments. It has been suggested that performers may integrate these sources of information in an approximately Bayesian fashion, by weighting available information sources according to their expected precision. In the present work, we extended Bayesian brain approaches to anticipation by using formal computational models to estimate how performers weighted different information sources when anticipating the bounce direction of a rugby ball. Both recreational (novice) and professional (expert) rugby players (n = 58) were asked to predict the bounce height of an oncoming rugby ball in a temporal occlusion paradigm. A computational model, based on a partially observable Markov decision process, was fitted to observed responses to estimate participants' weighting of online sensory cues and prior beliefs about ball bounce height. The results showed that experts were more sensitive to online sensory information, but that neither experts nor novices relied heavily on prior beliefs about ball trajectories in this task. Experts, but not novices, were observed to down-weight priors in their anticipatory decisions as later and more precise visual cues emerged, as predicted by Bayesian and active inference accounts of perception.

Viewpoint planning with transition management for active object recognition

Active object recognition (AOR) provides a paradigm where an agent can capture additional evidence by purposefully changing its viewpoint to improve the quality of recognition. One of the most concerned problems in AOR is viewpoint planning (VP) which refers to developing a policy to determine the next viewpoints of the agent. A research trend is to solve the VP problem with reinforcement learning, namely to use the viewpoint transitions explored by the agent to train the VP policy. However, most research discards the trained transitions, which may lead to an inefficient use of the explored transitions. To solve this challenge, we present a novel VP method with transition management based on reinforcement learning, which can reuse the explored viewpoint transitions. To be specific, a learning framework of the VP policy is first established via the deterministic policy gradient theory, which provides an opportunity to reuse the explored transitions. Then, we design a scheme of viewpoint transition management that can store the explored transitions and decide which transitions are used for the policy learning. Finally, within the framework, we develop an algorithm based on twin delayed deep deterministic policy gradient and the designed scheme to train the VP policy. Experiments on the public and challenging dataset GERMS show the effectiveness of our method in comparison with several competing approaches.

Task-evoked pupillary responses track precision-weighted prediction errors and learning rate during interceptive visuomotor actions

In this study, we examined the relationship between physiological encoding of surprise and the learning of anticipatory eye movements. Active inference portrays perception and action as interconnected inference processes, driven by the imperative to minimise the surprise of sensory observations. To examine this characterisation of oculomotor learning during a hand-eye coordination task, we tested whether anticipatory eye movements were updated in accordance with Bayesian principles and whether trial-by-trial learning rates tracked pupil dilation as a marker of 'surprise'. Forty-four participants completed an interception task in immersive virtual reality that required them to hit bouncing balls that had either expected or unexpected bounce profiles. We recorded anticipatory eye movements known to index participants' beliefs about likely ball bounce trajectories. By fitting a hierarchical Bayesian inference model to the trial-wise trajectories of these predictive eye movements, we were able to estimate each individual's expectations about bounce trajectories, rates of belief updating, and precision-weighted prediction errors. We found that the task-evoked pupil response tracked prediction errors and learning rates but not beliefs about ball bounciness or environmental volatility. These findings are partially consistent with active inference accounts and shed light on how encoding of surprise may shape the control of action.

Reclaiming saliency: Rhythmic precision-modulated action and perception

Computational models of visual attention in artificial intelligence and robotics have been inspired by the concept of a saliency map. These models account for the mutual information between the (current) visual information and its estimated causes. However, they fail to consider the circular causality between perception and action. In other words, they do not consider where to sample next, given current beliefs. Here, we reclaim salience as an active inference process that relies on two basic principles: uncertainty minimization and rhythmic scheduling. For this, we make a distinction between attention and salience. Briefly, we associate attention with precision control, i.e., the confidence with which beliefs can be updated given sampled sensory data, and salience with uncertainty minimization that underwrites the selection of future sensory data. Using this, we propose a new account of attention based on rhythmic precision-modulation and discuss its potential in robotics, providing numerical experiments that showcase its advantages for state and noise estimation, system identification and action selection for informative path planning.

Embodied Object Representation Learning and Recognition

Scene understanding and decomposition is a crucial challenge for intelligent systems, whether it is for object manipulation, navigation, or any other task. Although current machine and deep learning approaches for object detection and classification obtain high accuracy, they typically do not leverage interaction with the world and are limited to a set of objects seen during training. Humans on the other hand learn to recognize and classify different objects by actively engaging with them on first encounter. Moreover, recent theories in neuroscience suggest that cortical columns in the neocortex play an important role in this process, by building predictive models about objects in their reference frame. In this article, we present an enactive embodied agent that implements such a generative model for object interaction. For each object category, our system instantiates a deep neural network, called Cortical Column Network (CCN), that represents the object in its own reference frame by learning a generative model that predicts the expected transform in pixel space, given an action. The model parameters are optimized through the active inference paradigm, i.e., the minimization of variational free energy. When provided with a visual observation, an ensemble of CCNs each vote on their belief of observing that specific object category, yielding a potential object classification. In case the likelihood on the selected category is too low, the object is detected as an unknown category, and the agent has the ability to instantiate a novel CCN for this category. We validate our system in an simulated environment, where it needs to learn to discern multiple objects from the YCB dataset. We show that classification accuracy improves as an embodied agent can gather more evidence, and that it is able to learn about novel, previously unseen objects. Finally, we show that an agent driven through active inference can choose their actions to reach a preferred observation.

Posture Instability Is Associated with Dopamine Drop of Nigrostriatal System and Hypometabolism of Cerebral Cortex in Parkinson Disease

BACKGROUND

Posture instability (PI) is known to be a severe complication in Parkinson's disease (PD), and its mechanism remains poorly understood. Our study aims to explore the changes of brain network in PI of PD, and further investigate the role of peripheral inflammation on activities of different brain regions in PD with PI.

METHODS

167 individuals were recruited, including 36 PD cases with PI and 131 ones without PI. We carefully assessed the status of motor and cognitive function, measured serum inflammatory factors, and detected the dopaminergic pathways and the metabolism of different brain regions by positron emission tomography (PET). Data analysis was conducted by variance, univariate analysis, chi-square analysis, logistic regression, and partial correlation.

RESULT

No difference was found for age or onset age between the two groups (P>0.05). Female patients were susceptible to posture impairment and had a 2.14-fold risk for PI compared with male patients in PD (P<0.05). Patients with PI had more severe impairment of motor and cognitive function for a longer duration than those without PI (P<0.05). The mean uptake ratios of presynaptic vesicular monoamine transporter (VMAT2), which were detected in the caudate nucleus and putamen, were lower in PI group than those without PI (P<0.05). There were lower activities of the midbrain, caudate nucleus, and anterior medial temporal cortex in PI group than those in the non-PI group (P<0.05). Although serum concentrations of immunoglobulins (IgG, IgM, and IgA) and complements (C3, C4) were higher in PI group than those in the non-PI group, only serum IgM concentration had a significant difference between the two groups (P<0.05). We further explored significant inverse correlations of IgG, IgM, IgA, and C4 with activities of some cerebral cortex in PI of PD (P<0.05).

CONCLUSION

Female patients were susceptible to posture instability and had a 2.14-fold risk for PI of PD. Patients with PI had more severe impairments of motor and cognitive function for a longer duration than those without PI. PI was associated with dopamine drop of the nigrostriatal system and lower activities of the limbic cortex in PD. Peripheral inflammation may be involved in degeneration of the cerebral cortex in PD combined with PI.