Keynote Lectures

Abstract
Voice and voiced speech are biological signals produced by humans with communication purposes. Current signal processing methods allow the inversion of the voice producing apparatus to deepen into biomechanical systems driven by neuromotor signals. Complex inverse models may be designed and set-up to reconstruct semantically rich correlates which are related with organic as well as neurologic disease etiology. New detecting, tracking and monitoring tools and protocols may help in improving current protocols to help in the care of illnesses as Parkinson, Alzheimer or Lateral Amiotrophic Schlerosis, or in Voice Rehabilitation of the Laryngectomized, or in evaluating function restoration after Larynx Conservative Surgery, as well as in Singing Education and many other fields. The talk concentrates in discussing the basic methodologies and algorithms to focus on protocols, tools and results in the state-of-the-art.

Abstract
Source separation is a fundamental problem in signal processing. In early 80's, independent component analysis (ICA) have been developed for solving this problem, with very weak priors on sources hence the name "blind". ICA was then successfully applied in middle of 90 to biomedical signals, like ECG, EEG, MRI, etc. But, in fact, each class of  biomedical signals has usually known properties, which can be used as priors in source separation, and in addition (or instead) to source independence which is not always satisfied. Usual priors can be properties in frequency or time domains, model of sources. Using such priors, the source separation problem becomes semi-blind (or not blind at all !) and one can consider (or design) methods which are simpler and more efficient than ICA. In this talk, I first explain the main theoretical results for solving source separation using ICA or methods based on various priors. Then, I illustrate the various approaches proposed in the theoretical framework with a few examples in biomedical signal processing. The talk will finish with a discussion on a few open questions.

Abstract
The brain is a network of individual neurons, each transmitting electrical impulses and communicating via chemical signals. Monitoring the activity of these neurons is necessary to understand how the networks of the brain process sensory information and control behavior. There are many, many neurons in these networks (a cubic millimeter of brain tissue may contain ~100,000 neurons), yet current devices for recording their electrical activity monitor only a small fraction (~10 neurons per cubic millimeter). To progress our understanding of the function and pathology of the brain, we will require new devices capable of measuring neuron electrical activity at a network scale, as well as devices to detect the chemical signals that communicate and regulate this activity. I will discuss new approaches to build such devices, which borrow from advances in nano- and microfabrication, MEMS, and biosensors. While the application of these technologies to neuroscience holds great promise, many obstacles remain. New insights, from many different technical fields, will be required to produce the devices that allow us to transcend our current understanding of nature's most complex machine.

Abstract
Human movement involves a complex series of coordinated neural and muscloskeletal processes. A breakdown in any of these processes can result in abnormal movement. Movement Disorders collectively affect approximately 10 million people in Europe. Parkinson’s disease, essential tremor and primary dystonia are the three most common movement disorders in Europe. 500,000 people have some form of dystonia, of which 80,000 have primary dystonia. This talk will discuss results of recent behavioral and neuroimaging experiments to help better understand the pathogenesis and genetic basis and of dystonia. The talk will also discuss the analysis of movement recorded in realworld environments to provide a window into cognitive function.

Abstract
A main goal of multimodal interface now is to support natural, efficient, powerful, and flexible human-computer interaction for different types of virtual environments.  If the interaction technology is awkward, or constraining, the user’s experience with the synthetic environment is severely degraded. If the interaction itself draws attention to the technology, rather than the task at hand, it becomes an obstacle to a successful virtual environment interface. The traditional two-dimensional, keyboard- and mouse-oriented graphical user interface (GUI) is not well-suited for virtual environments. Instead, using several different modalities and integrating them provide the opportunity to develop user-friendly interface with a virtual environment. The cross product of communication modalities and sensing devices begets a wide range of multimodal interface techniques. The potential of these techniques to support natural and powerful interfaces is the future of virtual reality constructing and designing. To more fully support natural communication, it has to not only track human movement, but to interpret that movement in order to recognize semantically meaningful modality. While tracking user’s modalities may be quite useful to express them through higher-level relations such as distance, the relative direction of movement or orientation of the objects being tracked.

In this lecture, we shall consider the most popular approaches for capture, tracking and recognition of different modalities simultaneously to create intellectual human-computer interface for different goals. Taking into account the large gesture variability and their important role in creating intuitive interfaces, the considered approaches focus one's attention on gestures although the approaches may be used also for other modalities. The considered approaches are user independent and do not require large learning samples. In section 1 of the lecture gestures modalities are considered as natural and artificial. Before gestures recognition the parts of the body involved in gestures have to be captured in video stream.  Modern capture and tracking methods are included in section 2 of the lecture. If the part of the body has been captured as a digital image, it can be recognized using some mathematical recognition models. Section 3 of the lecture is devoted to the most effective recognition models. Multimodal aggregation as a way to an intellectual human-computer interaction is presented in section 4 of the lecture. The last section of the lecture is a conclusion.

Abstract
Systems medicine is an exciting new field which will provide effective answers to the challenges of integrating Big Data (meta analysis of datasets, multi omics, multi organs), characterize comorbidities, and building a multiscale model of human physiology.

Today multi-scale and complex biomedical data are gathered and analysed in a rather simple way that in many cases misses the opportunity to uncover combinations of predictive disease profiles; always the subtle associations about comorbidities found in the meta data analysis need bioinformatics to suggest the most appropriate experimental validation. Modeling is superior to the data-mining correlative approach in transforming data into knowledge but data still should be used for parameter estimation. This is particularly true both in computational systems biology and bioengineering where mechanistic models, based on deterministic and stochastic differential equations, could be devised from biological experimental knowledge.

Here the problem we encounter is of a different kind: we can observe what happens at almost all scales, from the whole organism down to the molecular level; however, putting things together in order to obtain real understanding is much more difficult and less developed.

One important aspect of multi scale modeling is the homogenisation of models across multi spatial scales, which allow cell-level models (using ODE or agent based) to be systematically scaled up to the tissue/organ level, and related asymptotic techniques for the analysis of multiple timescale problems.

Here the challenge is the model order reduction, i.e. to abandon high dimensional bioengineering systems in favour of the simplest mathematical model that “does the job”. Again the help comes from the bioinformatics analysis of the data which, leveraging on millions year evolution, describes the space of solutions explored by Nature. We discuss the powerful connection between bioinformatics and bioengineering through two case studies: the immune system response and the biomechanics of bone remodeling.  Both systems provide interesting examples of Big Data, comorbidities and multi scale systems.