Konstantinos Eftaxias

The 9th annual MLSP competition: New methods for acoustic classification of multiple simultaneous bird species in a noisy environment

Birds have been widely used as biological indicators for ecological research. They respond quickly to environmental changes and can be used to infer about other organisms (e.g., insects they feed on). Traditional methods for collecting data about birds involves costly human effort. A promising alternative is acoustic monitoring. There are many advantages to recording audio of birds compared to human surveys, including increased temporal and spatial resolution and extent, applicability in remote sites, reduced observer bias, and potentially lower cost. However, it is an open problem for signal processing and machine learning to reliably identify bird sounds in real-world audio data collected in an acoustic monitoring scenario. Some of the major challenges include multiple simultaneously vocalizing birds, other sources of non-bird sound (e.g., buzzing insects), and background noise like wind, rain, and motor vehicles.

An Informed Multitask Diffusion Adaptation Approach to Study Tremor in Parkinson's Disease

In this paper, a network-based approach for studying the relation between the tremor intensity and the brain connectivity of Parkinsons patients is introduced. We propose an adaptive multitask diffusion strategy to estimate the underlying model between the gait information and the electroencephalography signals. Furthermore, the method incorporates an S-transform-based connectivity measure that performs well even on a single-trial basis. The estimated connectivity values are then combined with the combination weights of the multitask diffusion strategy to model the relation between tremor and the brain signals. The outcome is an enhanced brain connectivity measure representing its time-space relation to the tremor. The results show how the differences between the connectivity values of patients with mild and severe hand tremor are most distinguishable when using the proposed method.

Challenges and trends in Ambient Assisted Living and intelligent tools for disabled and elderly people

In this review article, we present some challenges and opportunities in Ambient Assisted Living (AAL) for disabled and elderly people addressing various state of the art and recent approaches particularly in artificial intelligence, biomedical engineering, and body sensor networking.

Diffusion adaptive filtering for modelling brain responses to motor tasks

Diffusion adaptation combined with an adaptive way of estimating brain connectivity is used here in order to model specific motor tasks. We use Kalman filtering to fit an adaptive multivariate autoregressive model to our data and compute the connectivity measure which is a time-varying version of directed transfer function (DTF). The resulting method is used to classify the data from movement-related activities. The comparison between the proposed method and the non-diffusion method shows superiority of the former one.

Detection of Parkinson's tremor from EMG signals; a singular spectrum analysis approach

A robust constrained complex singular spectrum analysis approach for the assessment of Parkinsons tremor by separation of electromyograms (EMGs) is presented in this paper. This approach exploits the expected EMG characteristics of tremor within a subspace of the single channel surface EMG signal measured during the prescribed hand movement including flexion and extension and decomposed using singular spectrum analysis. The results are validated using the tremor signals simultaneously recorded using motion sensors.

Modelling brain cortical connectivity using diffusion adaptation

This work examines the flow of information among electrodes attached to the brain and uses diffusion adaptation strategies to assess brain cortical connectivity. The method uses the directed transfer function (DTF) technique to estimate combination coefficients to drive the adaptation and learning process. The diffusion strategy is then applied to the problem of recognizing left and right hand movements and its superior performance is demonstrated relative to solutions that rely on stand-alone electrodes and do not exploit coordination among multiple electrodes.

Joint EEG — EMG signal processing for identification of the mental tasks in patients with neurological diseases

Correlation size together with Lyapunov exponents estimated from both electroencephalography (EEG) and electromyography (EMG) signals, are the crucial variables in the classification of mental tasks using an artificial neural network (ANN) classifier for patients suffering from neurological disorders/diseases. The above parameters vary according to the status of the patient, for example: depending on how stressed or relaxed the patient is and what mental task is executed. The signals were acquired from patients with Parkinson disease, while they performed four different mental tasks. The performed mental states, detected with high specificity and accuracy, can help a completely paralyzed person (locked-in) to communicate with the environment through the brain waves, leading to increasing their quality of life.

The Ninth Annual MLSP Data Competition

The Ninth Annual Machine Learning for Signal Processing (MLSP) Data Competition Committee has hosted a bird classification challenge at Kaggle.com (http://www.kaggle.com/c/mlsp-2013-birds). For this years competition, participants were asked to develop classification algorithms to reliably identify the set of bird species in real-world audio data collected in an acoustic monitoring scenario. In this paper, we (the organizers of the competition) briefly describe the application, the data, the rules, and the outcomes of the competition. An MLSP record number of 79 teams entered the contest. We provided training data to the participants. The entries were tested using disjoint test data. The participants had access to the test data, but not the test labels. A separate multi-author long paper to summarize the new methods, which were contributed by multiple teams, will be included in this years conference proceedings. The two top ranking teams described their approaches in two separate companion papers, all of which will appear in this years conference proceedings. The first place team, whose entry produced an area under the receiver operating curve (ROC) of 0.956, is Gabor Fodor from Budapest University of Technology and Economics in Budapest, Hungary. The second place team, whose entry produced an area under the ROC of 0.951, consists of Hong Wei Ng and Thi Ngoc Tho Nguyen from Advanced Digital Sciences Center, University of Illinois at Urbana-Champaign, Singapore, Singapore.

Constrained LMS for dynamic flow networks

In this era of climate change, there is a growing need to offer adaptive learning algorithms in the optimisation of natural resources. These resources are typically optimised by evolutionary algorithms. However, evolutionary algorithms (EAs) are no longer adequate due to the ‘drift’ component introduced by environmental factors such as flash flooding. We therefore propose a novel constrained Least Mean Squares (LMS) algorithm for the optimisation of flow networks. For rigor, we provide a stability analysis of our adaptive algorithm, which enables us to interpret the physical meaning of the network at equilibrium. We evaluate our proposed method against genetic algorithm (GA), the most common evolutionary algorithm. The results are promising: not only the proposed constrained LMS has a performance advantage over GA, but its computational cost is significantly lower making it more suitable for real-time applications.

Cooperative learning for biomedical signal processing and recognition

Advances in adaptive signal processing and analysis of multichannel data recorded from human body, require cooperative learning to enable communications between the nodes of the network. In more complex cases where the human body is involved in executing multiple and different tasks simultaneously, the learning process is subject to both learning and differentiating between the sensor groups. In this article, the concept of connectivity, as applied to electroencephalograms (EEGs), followed by cooperative learning and the theory involved are explored. Next, the concept of multitask diffusion adaptation and learning will be discussed. Finally, we see how cooperative learning can model a complex biological system as well as restoration of movement related brain potentials. In addition, a cooperative tracking system is introduced for detection and tracking of the changes in brain event related potentials.