Visits to University of California, Berkeley, Stanford University, and SRI International

During his visit to the Department of Statistics, University of California, Berkeley, Prof. Cvetkovic will be hosted by Prof. Yu, in the Statistical Machine Learning group. He will be focusing on topics of the current activity of Prof. Yu's group, that can be broadly described as statistical machine learning theory, methodologies, and algorithms for solving high-dimensional data problems. Particular problems covered include sparse modelling (e.g. Lasso, compressed sensing), structured sparsity, analysis and methods for spectral clustering, and applications to data which come from a diverse range of interdisciplinary areas, ranging from neuroscience to social networks. During this visit, Prof. Cvetkovic and Prof. Yu will set forth directions for collaboration on problems in learning in high-dimensions, leading to a research grant proposal.

During his previous EPSRC project, EP/D053005/1, Prof. Cvetkovic in collaboration with Prof. Sollich, Department of Mathematics, King's College London, and Prof. Yu developed an unorthodox approach to robust speech recognition in high-dimensional spaces of acoustic waveforms of speech. Dr. Horacio Franco, Director of Speech Technology and Research Laboratory of SRI International, who in 2010 won a major DARPA award for solving the problem of the sensitivity of automatic speech recognition systems to additive noise, finds this approach groundbreaking and expresses a strong interest in exploring venues for collaboration.
The purpose of this visit would be to investigate ways in which the approach developed by Prof. Cvetkovic and his collaborators can be brought closer to practice and based on that investigate the directions of long-term collaboration and possible joint grant proposals between SRI International, King's College London, and UC Berkeley.

At King's College, Prof. Cvetkovic has commenced work on a new multichannel audio technology, supported by EPSRC grant EP/F001142/1. The project produced a considerable publication volume and patent portfolio. A visit to one of world leading centres for music and acoustics technologies, such as CCRMA, would be very beneficial for taking advantage of this gained momentum to penetrate the field, which is still a new application area for Prof. Cvetkovic, at a deeper level, expand its scope, establish collaborations, and inform future grant proposals. At CCRMA, Prof. Cvetkovic will be interacting primarily with Prof. Julius Smith, working on multichannel audio technologies, and other signal processing problems in audio and acoustics. A recent work of Prof. Cvetkovic complements a large volume of work of Prof. Smith on ultra fast rendition of multichannel audio using digital waveguide networks (DWNs). This is an area which is of a significant academic interest, requiring interdisciplinary approaches at the interface of signal processing, acoustics, psychoacoustic, and computer science, as well of a great relevance to virtual reality and gaming applications. While this would be the area of initial focus, at CCRMA there are several other ongoing projects which are closely related to Prof. Cvetkovic's research or research in the Institute of Telecommunications at King's (Mobile Phone Orchestra, Sound in Space, Music in Virtual Worlds), as well as projects which could provide valuable inspiration for possible collaborative projects between the Department of Music and the Institute of Telecommunications at King's and CCRMA (Sound Waves on the Internet for Real-time Echoes, and the Historical Recordings). Finally, most of the largest companies which are potential licensees of Prof. Cvetkovic'c audio technology, such as DTS, Dolby, Microsoft, are based on the west coast of the US. The presence of Prof. Cvetkovic at CCRMA would accelerate the exploration of licensing opportunities, as these and other relevant companies frequently visit CCRMA, and are situated in the Bay Area or not too far from it.

Visit to the Department of Statistics, University of California, Berkeley

The main purpose of the visit of Prof. Cvetkovic to the Department of Statistics, University of California, Berkeley, is to study emerging information sciences techniques and thus facilitate his engagement in cutting edge research in this field. His subsequent research in this field is meant to have a significant theoretical component, and immediate beneficiaries of this work would be other researchers in signal processing, and statistics and applied probability -- two areas which EPSRC intends to increase its investments in.

The techniques for statistical inference and kernel methods which Prof. Cvetkovic will be focusing on at Berkeley are applicable to a broad and diverse types of data, from social networks to neuroscience, so many segments of science, industry and society would benefit from this work. One particular application which will be considered is cardiac arrhythmia detection and classification. The ultimate benefit of this work would be improved healthcare and societal well being, as ventricular fibrillation (VF) is a leading cause of death in the western world and existing methods for VF detection are not sufficiently reliable. Other beneficiaries include pharmaceutical companies which develop medications for treating different forms of arrhythmias, then biomedical equipment companies which manufacture defibrillators, and finally other researchers in biomedical signal processing, and pharmacology.

Another application which will be considered is automatic speech recognition. The approach developed by Prof. Cvetkovic and Prof. Sollich in collaboration with Prof. Yu is still very novel and original, so results of this work will be of interest to the research community working on speech recognition. Beneficiaries of practical speech recognition systems built around these ideas are discussed in the following.

Visit to Speech Technology and Research (STAR) Laboratory, SRI International

Automatic speech recognition plays an important role in a wide variety of applications, ranging from collecting military intelligence, through assisted living and medical record transcription, to various customer service systems. Beneficiaries therefore include the military, healthcare, and service industries. Reliable and accurate automatic speech recognition systems contribute towards improving national security, providing better healthcare and reducing its cost, and making the IT infrastructure function seamlessly while appearing invisible. This work therefore addresses several important challenges within Digital Economy, Healthcare, and Global Uncertainties themes, as defined by EPSRC.

Visit to the Center for Computer Research in Music and Acoustics (CCRMA), Stanford University

The first and immediate line of beneficiaries of the work done during the visit to CCRMA are researches the fields of signal processing, music and acoustics technologies. Then companies producing systems for multichannel audio, including home stereo systems, game devices, and sound mixing consoles, make another line of direct beneficiaries. Through their products, people working in creative industries which involve sound recording, production and reproduction, will also benefit from this work. Finally, the ultimate beneficiary is the general public which will be able to enjoy superior audio quality at a lower price. Thus, the proposed activity will advance relevant science, it will lead to technological developments, which would have a positive impact on industry and commerce, and it will contribute towards enhancing the quality of life of general public.