Publications
233 results found
Garnelo M, Ricoult SG, Juncker D, et al., 2015, Variability and Reliabiltiy in Axon Growth Cone Navigation Decision Making, American Physical Society, Publisher: American Institute of Physics (AIP), ISSN: 0003-0503
Abbott WW, Thomik AA, Faisal AA, 2015, Sensory-motor system identification of active perception in ecologically valid environments, American Physical Society Annual meeting
The brain is a dynamical system mapping sensory inputs to motor actions. This relationship has been widely characterised by reductionist controlled experiments. Here we present work moving out of the lab ``into the wild'' to capture, rather than constrain, sensory inputs and motor outputs, by recording 90\% of sensory inputs using head mounted eye-tracking, scene camera and microphone as well as recording 95\% of skeletal motor outputs by motion tracking 51 degrees of freedom in the body and a total of 40 degrees of freedom from the hands. We can thus begin to systematically characterise the perception-action loop through system identification. This enables use to evaluate classical relationships in ecologically valid settings and behaviours including 3 daily scenarios: breakfast in the kitchen, evening chores and activities and in-door ambulation . This level of data richness (97 DOF, 60Hz), coupled with the extensive recordings of natural perceptual and behavioural data (total > 30 hrs, 10 subjects) enables us to answer general questions of how lab tasks and protocols will produce systematically different results from those found in daily life.
, 2015, Brain Informatics and Health - 8th International Conference, BIH 2015, London, UK, August 30 - September 2, 2015. Proceedings, Publisher: Springer, ISBN: 978-3-319-23343-7
Mehraban Pour Behbahani F, Taunton R, Thomik AAC, et al., 2015, Haptic SLAM for context-aware robotic hand prosthetics – simultaneous inference of hand pose and object shape using particle filters, 7th International IEEE EMBS Conference on Neural Engineering, Publisher: IEEE, Pages: 719-722
Even without visual feedback, humans can accurately determine the shape of objects on the basis of haptic feedback. This feat is achievable despite large variability in sensory and motor uncertainty in estimation of hand pose and object location. In contrast, most neuroprosthetic hands still operate unaware of the shape of the object they are manipulating and can thus only provide limited intelligence for natural control of the hand. We present a computational model for haptic exploration and shape reconstruction derived from mobile robotics: simultaneous localisation and mapping (SLAM). This approach solely relies on the knowledge of object contacts on the end-points, noisy sensory readings and motor control signals. We present a proof-of-principle accurate reconstruction of object shape (e.g. Rubik's cube) from single-finger exploration and propose a straightforward extension to a full hand model with realistic mechanical properties. The proposed framework allows for principled study of natural human haptic exploration and context-aware prosthetics. In conjunction with tactile-enabled prostheses, this could allow for online object recognition and pose adaptation for more natural prosthetic control.
Lorenz R, Monti R, Cole J, et al., 2015, Towards steering the chronnectome - on the potential of dynamic functional connectivity-based neurofeedback of large scale brain networks, Real-time Functional Imaging and Neurofeedback Conference
Abbott W, Thomik A, Faisal A, 2015, Embodied salience for gaze analysis in ecologically valid environments., J Vis, Vol: 15
The brain is a dynamical system, mapping sensory inputs to motor actions. This relationship has been widely characterised by reductionist controlled lab experiments. However, with the emergence of mobile eye-tracking, increasing emphasis has been placed on the ecological validity of gaze studies, taking them out of the lab and into the "wild" (Hayhoe & Ballard, 2005; Kingstone et al., 2003; Land & Tatler, 2009). Here we build on this by capturing, rather than constraining, sensory inputs and motor outputs in natural behaviour. We record 90% of sensory inputs using head mounted eye-tracking, scene camera and microphone. Simultaneously, recording 95% of skeletal motor outputs by motion tracking 51 degrees of freedom in the body and a total of 40 degrees of freedom in the hands. All tracking equipment is markerless and thus allows unconstrained behavioural monitoring "in the wild". The eye-tracker data is processed post-hoc to give 3D gaze position relative to the subjects' head and limb endpoints using our GT3D decoding method (Abbott & Faisal, 2011; Abbott & Faisal, 2012) and the motion capture data. This enables us to evaluate classical relationships in ecologically valid environments including 3 daily scenarios: breakfast in the kitchen, evening activities in the home and in-door ambulation. We find that the classical categorisation of gaze data to saccades and fixations is insufficient to capture eye-movement repertoire in natural behaviour. We spend the majority of daily life making smooth eye-movements directly coupled to body movements (eg VOR). Classically, allocation of gaze has been attributed to both bottom up scene salience and top down ongoing task demands. We propose a new method for analysing and interpreting eye-movements in the wild, by relating them directly to body posture. Thus, the interpretation of gaze and attention is integrated, not disembodied, from the kinematics of motor behaviour. Meeting abstract presen
Mehraban Pour Behbahani F, Faisal AA, 2015, The emergence of decision boundaries is predicted by second order statistics of stimuli in visual categorization, Vision Sciences Society (VSS)
Categorization is a fundamental and innate ability of our brain, however,its underlying mechanism is not well understood. Previous experimentalwork on human categorization shows data that is consistent with both discriminative and generative (typically Bayesian) classification. However,the experimental designs used were not able to deliver a test that unambiguously accepts one method while simultaneously rejecting the other. Therefore, we designed a novel experiment in which subjects are trained to distinguish two classes A and B of visual objects, while exemplars of each class are drawn from Gaussian parameter distributions, with equal variance and different means. During the experiment, we test how the subject’s representation of the categories changes as a result of being exposed to outliers for only one of the categories, A, far from category B (i.e. increasing category A’s variance). Generative classifiers are by necessity sensitive to novel information becoming available during training, which updates beliefs regarding the generating distribution of each class and assumes class A’s variance has increased significantly, thereby reaching across the region occupied by B which predicts an emergence of a new boundary. In contrast, discriminative classifiers are sensitive to novel information only if it affects the immediate discrimination of classes. We observe that, initially, when both categories have equal variance, subjects’ decision boundary lies between the two categories consistent with both discriminative and generative algorithms. However, the introduction of the outliers for category A, influences the subject’s knowledge of the distribution associated with alternative categories such that objects closer to category B and farthest from category A’s outliers will be classified as belonging to category A and thus a new boundary emerges, only predicted by our simulations of generative classifiers. These results give evidenc
Faisal A, Peltonen J, Georgii E, et al., 2014, Toward Computational Cumulative Biology by Combining Models of Biological Datasets, PLOS ONE, Vol: 9, ISSN: 1932-6203
- Author Web Link
- Cite
- Citations: 5
Neishabouri A, 2014, Saltatory conduction in unmyelinated axons: clustering of Na channels on lipid rafts enables micro-saltatory conduction in C-fibers, Frontiers in Neuroanatomy, ISSN: 1662-5129
The action potential (AP), the fundamental signal of the nervous system, is carried by two types of axons: unmyelinated and myelinated fibers. In the former the action potential propagates continuously along the axon as established in large-diameter fibers. In the latter axons the AP jumps along the nodes of Ranvier—discrete, anatomically specialized regions which contain very high densities of sodium ion channels. Therefore, saltatory conduction is thought as the hallmark of myelinated axons, which enables faster and more reliable propagation of signals than in unmyelinated axons of same outer diameter. Recent molecular anatomy showed that in C-fibers, the very thin (0.1 μm diameter) axons of the peripheral nervous system, Nav1.8 channels are clustered together on lipid rafts that float in the cell membrane. This localized concentration of Na+ channels resembles in structure the ion channel organization at the nodes of Ranvier, yet it is currently unknown whether this translates into an equivalent phenomenon of saltatory conduction or related-functional benefits and efficiencies. Therefore, we modeled biophysically realistic unmyelinated axons with both conventional and lipid-raft based organization of Na+ channels. We find that APs are reliably conducted in a micro-saltatory fashion along lipid rafts. Comparing APs in unmyelinated fibers with and without lipid rafts did not reveal any significant difference in either the metabolic cost or AP propagation velocity. By investigating the efficiency of AP propagation over Nav1.8 channels, we find however that the specific inactivation properties of these channels significantly increase the metabolic cost of signaling in C-fibers.
Kotti M, Duffell LD, Faisal AA, et al., 2014, The Complexity of Human Walking: A Knee Osteoarthritis Study, PLOS ONE, Vol: 9, ISSN: 1932-6203
- Author Web Link
- Open Access Link
- Cite
- Citations: 18
Kotti M, Duffell LD, Faisal AA, et al., 2014, Towards predicting the effectiveness of knee surgery for knee osteoarthritis patients, Young Researchers’ Futures Meeting: Engineering for Orthopaedic Applications
Osteoarthritis (OA) is the commonest form of musculoskeletal disability. Surgery is usually used to manage the end stage of the disease taking the form of either total joint or unicompartmental replacement. The outcome of such surgeries however could be disappointing. This work aims to exploit machine learning [1] to predict the effectiveness of surgery with respect to return to normal activities, as assessed using the Tegner activity score.
Kotti M, Duffell LD, Faisal AA, et al., 2014, Analysis of knee osteoarthritis ground reaction vertical force during stair ascent: A neural network approach, 7th World Congress on Biomechanics
Osteoarthritis (OA) is the second leading cause of pain and disability, affecting more than 250 million people worldwide. Here, we focus on knee OA, the most common form of OA. To collect data, subjects were asked to ascend a custom-based stair with a force plate (Kistler Type 9286B, Kistler Instrumente AG, Winterthur, Switzerland). Each subject was barefoot and provided 3 trials. We consider the strike of the right foot on the force plate. Trial data where subjects did not cleanly strike the force plate was excluded from the analysis, so a total of 272 trials were recorded. The signal from the force plate was recorded at a sampling rate of 1000 Hz, then normalised to the subject’s body weight (N/kg), and time-normalised to the entiregait cycle using linear interpolation. We retain the ground reaction force over the vertical plane. Out of the 96 subjects, 37 have OA at the one knee, 11 at both knees and the remaining 48 are control subjects. To automatically classify the motion patterns into three categories, i.e. normal, knee OA at one knee, and knee OA at both knees, a probabilistic neural network (PNN) was employed. The PNN is based on the theory of Bayesian classification. Regarding the PNN structure, it is a feed-forward neural network with high degree of parallelism. The PNN classifier is a non-parametric classification approach, since we have no guarantee that the data follows a Gaussian distribution. The effectiveness of the PNN for detecting knee OAhas been verified previously, but the data analysed were radiographic images, rather that ground reaction forces. Results for 5-fold cross-validation can be seen in the Table below. To conclude, the PNN is able to effectively handle locomotion data that exploit great variability both inter- and intra-subject. Also, the PNN can detect approximately 16% of subjects that claim not to have knee OA, but they present gait patterns similar to those of subjects that suffer knee OA.
Neishabouri A, Faisal AA, 2014, Axonal Noise as a Source of Synaptic Variability, PLOS COMPUTATIONAL BIOLOGY, Vol: 10, ISSN: 1553-734X
- Author Web Link
- Open Access Link
- Cite
- Citations: 25
Rodríguez M, Sylaidi A, Faisal AA, 2014, Developing a novel fMRI-Compatible motion tracking system for haptic motor control experiments, Pages: 25-30
Human neuroimaging can play a key role in addressing open questions in motor neuroscience and embodied cognition by linking human movement experiments and motor psychophysics to the neural foundation of motor control. To this end we designed and built fMOVE, an fMRI-compatible motion tracking system that captures 3DOF goal-directed movements of human subjects within a neuroimaging scanner. fMOVE constitutes an ultra-low-cost technology, based on a zoom lens high-frame rate USB camera and, our adaptation library for camera-based motion tracking and experiment control. Our motion tracking algorithm tracks the position of markers attached to a hand-held object. The system enables to provide the scanned subjects a closed-loop real time visual feedback of their motion and control of complex, goal-oriented movements. The latter are instructed by simple speed-accuracy tasks or goal-oriented object manipulation. The system's tracking precision was tested and found within its operational parameters comparable to the performance levels of a scientific grade electromagnetic motion tracking system. fMOVE thus offers a lowcost methodological platform to re-approach the objectives of motor neuroscience by enabling ecologically more valid motor tasks in neuroimaging studies.
Lourenço PR, Abbott WW, Faisal AA, 2014, EEG and eye-tracking integration for ocular artefact correction, Pages: 79-86
Electroencephalograms (EEG) are a widely used brain signal recording technique. The information conveyed in these recordings can be an extremely useful tool in the diagnosis of some diseases and disturbances, as well as in the development of non-invasive Brain-Machine Interfaces (BMI). However, the non-invasive electrical recording setup comes with two major downsides, a. poor signal-to-noise ratio and b. the vulnerability to any external and internal noise sources. One of the main sources of artefacts are eye movements due to the electric dipole between the cornea and the retina. We have previously proposed that monitoring eye-movements provide a complementary signal for BMIs. He we propose a novel technique to remove eye-related artefacts from the EEG recordings. We couple Eye Tracking with EEG allowing us to independently measure when ocular artefact events occur and thus clean them up in a targeted manner instead of using a "blind" artefact clean up correction technique. Three standard methods of artefact correction were applied in an event-driven, supervised manner: 1. Independent Components Analysis (ICA), 2. Wiener Filter and 3. Wavelet Decomposition and compared to "blind" unsupervised ICA clean up. These are standard artefact correction approaches implemented in many toolboxes and experimental EEG systems and could easily be applied by their users in an event-driven manner. Already the qualitative inspection of the clean up traces show that the simple targeted artefact event-driven clean up outperforms the traditional "blind" clean up approaches. We conclude that this justifies the small extra effort of performing simultaneous eye tracking with any EEG recording to enable simple, but targeted, automatic artefact removal that preserves more of the original signal.
Lorenz R, Faisal AA, Dinov M, et al., 2014, Neurofeedback training of large-scale brain networks, Annual Meeting of Society of Neuroscience
Mehraban Pour Behbahani F, Faisal AA, 2014, Visual categorization reflects second order generative statistics, Cosyne
Categorization is a fundamental and innate ability of our brain, however, its underlying implementation is not well understood. Computationally, the task of classification is the same for humans as for machines. In machine learning, classification algorithms fall broadly into Discriminative and Generative (typically Bayesian) algorithms. Previous experimental work on human categorization shows data that is consistent with both discriminative and generative classification. However, the experimental designs used were not able to capture a desired test that unambiguously accepts one method while simultaneously rejecting the other. Therefore, we designed a novel experiment in which subjects are trained to distinguish two classes A and B of visual objects, while exemplars of each class are drawn from Gaussian parameter distributions, with equal variance and different means. During the experiment, we test how the subject’s representation of the categories change as a result of being exposed to outliers for only one of the categories, A, far from category B (i.e. increasing category A’s variance). Generative classifiers are by necessity sensitive to novel information becoming available during training, which updates beliefs regarding the generating distribution of each class. In contrast discriminative classifiers are sensitive to novel information only if it affects the immediate discrimination of classes. Our results show that, initially, when both categories have equal variance, subjects’ classification behavior is consistent with both discriminative and generative algorithms. However, the introduction of the outliers for category A, influences the subject’s knowledge of the distribution associated with alternative categories such that objects closer to category B and farthest from category A’s outliers will be classified as belonging to category A, only predicted by our simulations of generative classifiers. This confirms and extends our p
Mehraban Pour Behbahani F, Faisal AA, 2014, Human Visual categorization is only consistent with Bayesian generative representations, Society for Neuroscience (SfN)
In neuroscience, the generative framework of Bayesian Decision Theory has emerged as a principled way to predict how the brain has to act in the face of uncertainty (Ernst & Banks, 2002, Körding & Wolpert, 2004, Faisal et al., 2008). We hypothesize that the brain might also use generative Bayesian principles to implement its categorization strategy. Previous experimental work on human categorization shows data that is consistent with both discriminative and generative classification (Hsu & Griffiths, 2010) and did not allow confirming the implementation of one or the other. Therefore, we designed a novel experiment in which subjects are trained to distinguish two classes A and B of visual objects drawn from Gaussian parameter distributions with equal variance. During two different experimental paradigms, we test how the subject’s representation of the categories change after being exposed to outliers for only one of the categories, A, far from category B. Generative classifiers are by necessity sensitive to novel information becoming available during training, which updates beliefs regarding the generating distribution of each class. In contrast, discriminative classifiers are sensitive to novel information only if it affects the immediate discrimination of classes. In the first paradigm, we characterize the categorization boundary between the two classes and track the shift in the boundary after the introduction of outliers. A generative classifier will prompt to reconsider the variance of class A and shifts the boundary towards category B accordingly. However, the discriminative classifier will not react as there is no new information added to the boundary itself. Our second paradigm provides an even more stringent test for generative models: again, outliers for class A are presented far away from class B. Additionally, the two classes are selected to be close enough, such that a generative classifier would assume that class A’s varian
Gavriel C, Faisal A, 2014, A Comparison of Day-Long Recording Stability and Muscle Force Prediction between BSN-Based Mechanomyography and Electromyography, Wearable and Implantable Body Sensor Networks (BSN), 2014 11th International Conference on, Pages: 69-74
Day-long continuous monitoring requires stable sensors that can minimise the effects of drift and maintain high accuracy and precision over time. We have recently shown that our inertial motion tracking system can capture stable kinematic data, calibrated against ground-truth over a long period of time. However, for many clinical and daily life activities, it is also essential to monitor the muscle-activity. In this study, we evaluate the long-term recording stability of our prototype mechanomyography (MMG) sensors as an extension to our existing ETHO1 body sensor network platform. We attached the MMG sensors along with commercial high-accuracy EMG electrodes on the arm muscles of 5 subjects throughout a working day of 9 hours. The subjects followed their daily routine but they had to perform a multi-level force-matching task through flexion and extension of their arm during four short sessions of the day, as measures of practical signal quality. We designed a force predictor that used either EMG or MMG signals to predict the forces generated by subjects. Our prototype low-cost MMG channels have shown comparable results (RMSE: 23N and R2: 0.91) in predicting the force levels applied when compared against the commercial high-accuracy EMG sensor (RMSE: 19N and R2: 0.95).
Haber D, Thomik AAC, Faisal A, 2014, Unsupervised Time Series Segmentation for High-Dimensional Body Sensor Network Data Streams, IEEE/EMBS Body Sensor Networks (BSN), Pages: 121-126
The vast amounts of data which can be collected using body-sensor networks with high temporal and spatial resolution require a novel analysis approach. In this context, state-of-the-art Bayesian approaches based on variational, non-parametric or MCMC derived methods often become computationally intractable when faced with several million data points. Here, we present how the simple combination of PCA, approximate Bayesian segmentation and temporal correlation processing can achieve reliable time series segmentation. We use our method, which relies on simple iterative covariance, correlation and maximum likelihood operations, to perform complex behavioural time series segmentation over millions of samples in 18 dimensions in linear time and space. Our approach is suitable for even higher dimensional data streams as performance scales near constantly with the dimensionality of the time series samples. We validate this novel approach on an artificially-generated time series and demonstrate that our method is very robust to noise and achieves a segmentation accuracy of over 86% of matching segments against ground-truth. We conclude that our approach makes Big Data driven approaches to stream processing Body Sensor Network (BSN) data tractable, and is required for BSN-driven Neurotechnology applications in Brain-Machine Interfacing and Neuroprosthetics.
Hecht EE, Gutman DA, Khreisheh N, et al., 2014, Acquisition of Paleolithic toolmaking abilities involves structural remodeling to inferior frontoparietal regions, Brain Structure and Function, Pages: 1-17, ISSN: 1863-2653
Fara S, Gavriel C, Vikram CS, et al., 2014, Prediction of Arm End-Point Force Using Multi-channel MMG, Wearable and Implantable Body Sensor Networks (BSN), 2014 11th International Conference on, Pages: 27-32
We investigate the effectiveness of a dual-channel MMG signal recorded from the biceps and triceps brachii as a way to predict the isometric forces produced by flexion and extension of the elbow. We asked 8 subjects to apply a range of isometric force levels for both flexion and extension of the elbow while the activity of the two muscles was captured using custom-built MMG sensors. By extracting two characteristic MMG features, the 'MMG score' and the root mean square power spectrum (rmsPS), we applied an artificial feed-forward neural network (NN) to generate a mapping between the MMG signals and the actual forces generated. The accuracy of the NN predictor was evaluated using a 10-fold cross validation, achieving an average across subject R2 of 0.76 and a RMSE of 8.6% of the maximum voluntary isometric contraction (MVC).
Bothe MK, Dickens L, Reichel K, et al., 2013, The use of reinforcement learning algorithms to meet the challenges of an artificial pancreas, EXPERT REVIEW OF MEDICAL DEVICES, Vol: 10, Pages: 661-673, ISSN: 1743-4440
- Author Web Link
- Cite
- Citations: 53
Kotti M, Duffell LD, Faisal AA, et al., 2013, Detecting knee osteoarthritis with statistical and machine learning approaches, Ascot, UK, MEC Annual Meeting 2013
Kotti M, Duffell LD, Faisal AA, et al., 2013, Analysis of knee osteoarthritis data via bioinformatics tools, Cardiff, UK, iEOS2013
Kotti M, Duffell LD, Faisal AA, et al., 2013, A PPCA-Bayesian approach for knee osteoarthritis classification using ground reaction reaction forces, ESB 2013, Publisher: European Society of Biomechanics
Kotti M, Duffell LD, Faisal AA, et al., 2013, Towards automatically assessing osteoarthritis severity by regression trees & SVMs, Natal, Brazil, ISB 2013
Abbott WW, Faisal AA, 2013, Embodied attention for gaze analysis in daily life activities., First International Workshop on Solutions for Automatic Gaze Data Analysis 2013 (SAGA 2013)
Abbott WW, Faisal AA, 2013, Ultra-low cost 3D gaze tracking: the non-invasive approach to neuroprosthetics, European Conference on Eye Movements (ECEM)
Abbott WW, Faisal AA, 2013, Using ultra-low cost 3D gaze tracking as an intuitive, non-invasive, high information throughput alternative to BMIs., Bernstein Compututational Neuroscience Conference
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.