Week 1. System level understanding of the brain and its emulation in advanced technology.
Activation patterns across focal seizures in patients with Epilepsy.
Rodrigo Rocamora Epilepsy Unit Hospital del Mar, Barcelona, Spain
The type of patient we attend in the Epilepsy Unit of the Hospital del Mar are people with resistance to drug treatment and who are candidates for surgical treatment. In our epilepsy unit the brain activity of patients with highly complex epilepsies, and that do not have brain lesions, is monitored for long periods of time thanks to an intracerebral electrode implantation system using a robot.
’Stroke, brain networks, and behavior’.
Maurizio Corbetta Padua Neuroscience Center, IT
Novel techniques for manipulating memory in humans.
Bryan A. Strange Laboratory for Clinical Neuroscience, Centre for Biomedical Technology, Technical University of Madrid
Memory for events and episodes can be impaired in a wide range of neurological and psychiatric diseases. With the incidence of dementia rising, there is an increasing need for techniques that can improve episodic memory in humans. By contrast, emotionally traumatic memories are central to the development of certain psychiatric conditions, such as post-traumatic stress disorder, necessitating treatments that weaken pre-existing, maladaptive or unwanted, memories. I will explain how an understanding of the neurobiological mechanisms underlying episodic memory has led us to develop methods to improve, or deliberately impair, memory in humans via the application of behavioural or pharmacological interventions, deep-brain stimulation, or electro- convulsive therapy. With a view towards identifying people who will require help to maintain memory function, I will lastly describe data from a large (n=1213), longitudinal cohort of elderly individuals presenting the neuropsychological and MRI biomarkers that predict the development of future mild-cognitive impairment in cognitively healthy people.
Application of neuroimaging to improve System-level understanding of brain pathology underlying neurological and psychiatric diseases.
Juergen Dukart Research Center Jülich, INM-7, Brain and Behaviour, DE
In my talk, I will present how multimodal functional and structural neuroimaging are applied to improve our system-level understanding of brain pathology in different neurological and psychiatric diseases. I will further illustrate how this system-level knowledge can be applied to improve diagnosis, prognosis and treatment across those indications.
The Virtual Brain: A neuroinformatics platform for ppersonalized brain simulation.
Petra Ritter Charité Universitätsmedizin Berlin, DE
The focus of this talk is personalized brain simulation. We have joined forces with international partners to create the open-source neuroinformatics platform The Virtual Brain. To produce the simulations, we are integrating data gathered from patients who have, for example, experienced a stroke or developed a brain tumor. The result is a personalized brain model for every patient, which contributes towards a better understanding of disease processes in the brain and enables the development of new therapy approaches. The principle is simple: the more data we are able to integrate, the more precise the model becomes. The vision for the future is that the treatment of brain disorders will be planned using a digital copy of the person’s brain. The Virtual Brain also allows to simulate possible scenarios of disease progression, potentially allowing future treatment plans to be modified early on. The Virtual Brain comprises also a central project in the EU Flagship Human Brain Project and is at the core of the consortium VirtualBrainCloud where 17 European Research Organizations, Patient Organizations and enterprises work together to create a cloud-based brain simulation platform for neurodegenerative disease.
Brain-inspired decision making at the system level.
Frederic Alexandre Institute for Research in Computer Science and Automation, FR
Decision making is one of the most fundamental cognitive functions for intelligent behavior. It is also one of the most complex, since it can be done in very different ways, from habitual behavior to explicit deliberations and integrating a variety of cognitive resources. For this reason and also considering our ability to flexibly switch from one form of decision process to another, a system level approach is recommended when studying brain circuits responsible for decision making. In this talk, I will introduce the corresponding brain architecture and present some recent work in my lab on this topic.
The role of cortico-basal ganglia-thalamo-cortical loops for working memory, categorisation and habitual learning: A neuro-computational approach towards cognition.
Fred Hamker Technical University Chemnitz, DE
The basal ganglia form multiple loops with cortex which are involved in cognitive processes and motor control. I will demonstrate examples of how the basal ganglia may be involved in cognitive processes such as working memory, consciousness, categorisation and habitual learning. Our results support the hypothesis, that the BG trains the PFC to acquire category knowledge. Due to its fast learning the BG can quickly find a good solution and further train the PFC to reach a high generalization performance. Finally, I address multiple cortex – basal ganglia loops and propose that cortex – basal ganglia loops are hierarchically organized. Each loop learns to select an intermediate objective at a different abstraction level, moving from goals in the ventral striatum to motor in the putamen. Habitual learning does not require any arbitration to control between different loops, but is explained by short-cuts between loops – a proposal not suggested before in the literature to explain habitual behavior.
Active efficient coding in vision: from robots to patients.
Jochen Triesch Frankfurt Institute for Advanced Studies.
Biological vision systems learn to perceive their environment largely autonomously, through interacting with it. How does this work? I will review recent work on building vision systems that autonomously learn a range of active vision skills based on principles of efficient coding. Interestingly, these systems reach sub-pixel accuracy without any supervision or kinematic model. At the same time, they are robust to various perturbations. Their behavior and learned representations mimic biological findings. Finally, I discuss how these models can also help us to better understand developmental disorders of vision in human patients.
Online adaptive coordination of model-based and model-free reinforcement learning systems in the brain and in robots.
Mehdi Khamassi Institute of Intelligent Systems and Robotics, FR
The reinforcement learning (RL) theory constitutes a computational framework to account for how animals adapt their actions by trial and error in order to maximize the amount of reward they get from the environment and minimize the amount of “punishments” (or aversive feedback) they get from it. In particular, this framework proposes a set of mathematical equations to describe how appropriate reinforcement signals which enable to efficiently adapt action selection should be computed under the form of reward prediction errors: if something better than expected occurs, this should lead to a positive reinforcement; if something worst than expected, there should be a negative reinforcement; and if something totally predicted occurs, there shouldn’t be any reinforcement because there is nothing to learn anymore.
Interestingly, this framework enables to well explain one type of behavioral “strategies” displayed by animals during instrumental learning tasks, Pavlovian conditioning tasks, or navigation tasks, whereby animals learn local action values (these values corresponding to the amount of expected future reward which could be obtained after performing these actions) in associations to stimuli or states of the environments (e.g., current location). The RL framework also enables to well describe the function subserved by dopaminergic neurons’ phasic responses to unpredicted rewards: they might encode some sort of reward prediction error (Schultz et al., 1997, Science).
Nevertheless, animals, and in particular mammals, also sometimes show different learning strategies which can be accounted for by extending the RL framework by considering that the animal learns an internal (mental) model of the structure of the environment (e.g., a cognitive map). We conventionally call these strategies ‘model-based’ as opposed to the previous ‘model-free’ strategies. Learning an internal model of the environment can be achieved through latent learning and enables faster adaptation to changes in the environment, at the cost of heavy computation cost and decision times.
In this presentation, I will give some intuitions about how these models work, with simple equations, and how they can account for behavioral and neurophysiological experimental data. I will show some examples of how a combination of model-free and model-based reinforcement learning permit to explain cases where animals alternate between goal-directed behaviors and habitual behaviors, the latter being slowly acquired through reward-based reinforcements and being inflexible. I will also show how developing a computational model to account for existing experimental data can then be tested in robots, in order to evaluate its robustness and efficiency in the real-world. In return, robotics’ test of neurobiological models can help raise novel model-driven predictions, leading to new experiments to test these predictions, and novel experimental findings.
The presentation will focus on examples of applications to navigation and instrumental conditioning experiments in rodents. In most cases, I will try to relate some of the model mechanisms to dopamine functions, as well as some neural activity in the prefrontal cortex, basal ganglia and hippocampus.
Together these results suggest a variety of mechanisms on which dopamine can impact, which can be progressively better understood through the tight collaboration between neurophysiological experimentation, computational modeling and neurorobotic validation.
Week2. Brain function, Motor learning, and Behavior.
The application of sensory error manipulations to motor rehabilitation and diagnostics.
Belen Rubio Ballester Institut for Bioengineering of Catalonia, ES
When exposed to visual error feedback, the motor system rapidly learns to adapt future motor commands and update outcome predictions thus optimizing motor planning and action selection. However, in cerebellar and stroke patients with proprioceptive impairments, motor adaptation may be altered. In this talk, I will present a set of studies supporting that the manipulation of visual error magnitudes during training may foster motor learning and promote recovery.
Klaudia Grechuta Institut for Bioengineering of Catalonia, ES
Practical, neuroscience-based rehabilitation.
Eiennne Burdet Imperial College, London, UK
Rehabilitation of upper limb motor function after stroke. A clinical perspective from highly functional to severely impaired individuals.
Roberto Llorens Universitat Politècnica de València, ES
Functional impairment of the upper limbs is a common sequelae after stroke that affects up to 85% of the survivors and persists, with a certain degree of severity, in 30 to 60% of the cases, six months after the onset. Given the incidence of upper limb deficits after stroke, and its
impact on the participation in activities of daily living, social life, and quality of life, rehabilitation is an imperative goal of physical and occupational therapy. A growing body of research shows potential benefits derived from the use of virtual reality and associated
technologies that enable the provision of tailored experiences to promote the recovery of motor function after stroke. However, most of the efforts have focused on mild to moderate impairments, leaving highly functional and severely impaired individuals out of the scope. This
talk will provide several clinical approaches that try to fill this gap.
Brain plasticity and new intervention strategies for stroke.
Adrian G. Guggisberg University of Geneva, CH
Advances in methods for assessing the human brain network have provided new insights into the consequences of stroke and into mechanisms underlying recovery. Thereby, it has become clear that stroke has impact on the entire brain and its network properties and can therefore be considered as a network disease. Neurological deficits do not only arise from focal tissue damage but also from a pathological brain state with altered white-matter tracts and disrupted neural interactions among wide-spread networks. Similarly, learning and clinical improvements are associated with specific compensatory structural and functional patterns of neural network interactions. Innovative treatment approaches such as neurofeedback and non-invasive brain stimulation have been successful in targeting such network patterns to enhance recovery. Network assessments show promise for predicting treatment response and for individualizing rehabilitation.
Wearable rehabilitation robots as tools for promoting voluntary movement function.
Juan C Moreno Instituto Cajal, CSIC, Madrid, ES
People who suffer a stroke experience weakness, postural instability, and spasticity that impair function and, particularly for more severely affected individuals, restricted mobility. Research efforts in neurorehabilitation technologies have been directed towards creating robotic exoskeletons to improve medical care and treatment. In this context, a major limitation is the inability to evoke and enable voluntary control in neurologically impaired individuals, making current wearable rehabilitation robots not yet optimal to enhance motor function. Cooperative motor behavior engages specific areas of the motor system compared with the execution of a task alone. The talk will describe results from EU Projects that developed robotic platforms exploiting this idea. First, we will discuss how sensorization and visual feedback already can improve the efficacy of the manual conventional approach for treating ankle spasticity. Secondly, the talk will present one study to understand the effects of biofeedback content when used for robotic post-stroke gait rehabilitation. In the last part, two cases will describe a) how continuous voluntary control of wearable robotic exoskeletons can be possible when linked via a patient-specific modeling, and b) discuss the design of an autonomous control system for video game-based exercises to train control of ankle muscles.