Compressed sensing

Emmanuel Candes, Caltech One of the central tenets of signal processing and data acquisition is the Shannon/Nyquist sampling theory: the number of samples needed to capture a signal is dictated by its bandwidth. This tutorial surveys a novel sampling or sensing theory which goes somewhat against this conventional wisdom. This theory now known as ``Compressed Sensing'' or ``Compressive Sampling'' allows the faithful recovery of signals and images from what appear to be highly incomplete sets of data, i.e. from far fewer measurements or data bits than traditional methods use. We will present the key ideas underlying this new sampling or sensing theory, and will survey some of the most important results. We will emphasize the practicality and the broad applicability of this technique, and discuss what we believe are far reaching implications; e.g. procedures for sensing and compressing data simultaneously and much faster. Finally, there are already many ongoing efforts to build a new generation of sensing devices based on compressed sensing and we will discuss remarkable recent progress in this area as well.

Signal processing for integrative bioinformatics

Al Hero, University of Michigan Bioinformatics will evolve to a point where it is possible to perform an integrated study of interacting biometric indicators in a population of organisms and use this information to accurately predict phenotypical behavior such as development, aging and disease. Information theory provides a natural framework for integration of diverse sources of data. This tutorial will cover elements of gene transcription and regulation, modalities and primary data sources for finding transcription factors, and information theoretic methods for reconstruction of transcriptional pathways. The Gata family of transcription factors affecting nephrogeneisis and kidney development will serve as our primary illustrative example throughout this tutorial.

Visual recognition: why, how, when

Pietro Perona, Caltech Our visual system is of great help as we carry on with our life. For example, we use it to recognize things (frogs, kitchens, tree bark, Elvis) quickly and reliably without touching them. Replicating this ability in machines would be very useful in a great number of applications: human-machine interfaces, surveillance, searching image and movie collections by content, manufacturing, self-driving vehicles. Visual recognition requires solving three problems: first, how to represent objects and categories with models that are specific, and yet accomodates the variability that is naturally found in nature (we recognize both poodles and corgies as `dogs'). Second, how to match models to images despite clutter, occlusion, lighting and viewpoint variations (I can find my shoes at the gym despite a great number of pieces of clothing strewn about the floor). The third problem is the most difficult: how can we acquire models in the first place: how can we become good at recognizing ice-cream cones, skateboards and sailing-boats when, at birth, we have no such knowledge? Thinking about visual recognition, and especially about visual learning, makes us reflect about fundamental issues in information engineering: how can we improve our knowledge of the world day-by-day without much external supervision, leveraging our current knowledge and creating new categories on the fly? How many training examples do we need to form new concepts? I will survey our theoretical and experimental understanding of visual recognition, which has much improved in the past 10 years, I will speculate on the progress we expect to make, and I will highlight a number of tantalizing open problems.

Multimedia retrieval: audio, speech, images and video

Malcolm Slaney, Yahoo Research Even more than text, the web has fostered an explosion of multimedia data. This wealth of data is both a treasure because it is so easy for people to create and much of it is quite compelling to watch. But it also strains the information retrieval world in ways that we don't fully understand. Yahoo has 25000 pictures of the Golden Gate Bridge in our databases. But users want just one image, or perhaps just one page of results. How can we find the most relevant images for people? In this talk I'd like to share with you Yahoo's experiences in this brave new world of multimedia everywhere, describe some promising technologies, and discuss open research directions.