Final year PhD student, University of Manchester
This Research aims to develop novel machine learning techniques for early detection of dementia
Alzheimer’s disease (AD) is a progressive, age-associated neurodegenerative disorder with a global prevalence likely to double every 20 years. There is no single test or biomarker that can detect or diagnose AD [Prince et al http://eprints.lse.ac.uk/67858/], yet predicting the earliest stages of cognitive impairment is vital to enable early intervention and possible disease modification. Currently, diagnosis relies on clinic-based assessment settings and diagnosis often occurs in the moderate or advanced stages of disease progression, making prevention or disease modification impossible.
Our aim is to develop novel data mining and analysis techniques based on a deep machine learning (DML) methodology, to ascertain early cognitive and functional decline indicative of degenerative brain disease from real life continuous, unobtrusively collected sensor data. DML techniques, which can detect abstract and complex patterns in large data sets, have demonstrated state-of-the-art performance in challenging artificial intelligence tasks, such as drug discovery [LeCun et al doi:10.1038/nature14539]. However, application of DML to early detection of AD from multimodal sensor data is not well studied.
Abstract: Deep learning is a popular machine learning approach which has achieved a lot of progress in all traditional machine learning areas. Internet of thing (IoT) and Smart City deployments are generating large amounts of time-series sensor data in need of analysis. Applying deep learning to these domains has been an important topic of research. The Long-Short Term Memory (LSTM) network has been proven to be well suited for dealing with and predicting important events with long intervals and delays in the time series. LTSM networks have the ability to maintain long-term memory. In an LTSM network, a stacked LSTM hidden layer also makes it possible to learn a high level temporal feature without the need of any fine tuning and preprocessing which would be required by other techniques. In this paper, we construct a long-short term memory (LSTM) recurrent neural network structure, use the normal time series training set to build the prediction model. And then we use the predicted error from the prediction model to construct a Gaussian naive Bayes model to detect whether the original sample is abnormal. This method is called LSTM-Gauss-NBayes for short. We use three real-world data sets, each of which involve long-term time-dependence or short-term time-dependence, even very weak time dependence. The experimental results show that LSTM-Gauss-NBayes is an effective and robust model.
Pub.: 13 Aug '17, Pinned: 25 Aug '17
Abstract: The function space of deep-learning machines is investigated by studying growth in the entropy of functions of a given error with respect to a reference function, realized by a deep-learning machine. Using physics-inspired methods we study both sparsely and densely-connected architectures to discover a layer-wise convergence of candidate functions, marked by a corresponding reduction in entropy when approaching the reference function, gain insight into the importance of having a large number of layers, and observe phase transitions as the error increases.
Pub.: 04 Aug '17, Pinned: 25 Aug '17
Abstract: This study investigated the application of a newly developed neuropsychological assessment, the Wolfenbütteler Dementia Test for Individuals with Intellectual Disabilities (WDTIM) in combination with the Dementia Screening Questionnaire for Individuals with Intellectual Disabilities (DSQIID).The instruments were evaluated in a prospective 2-year follow-up study. A total of 102 people with an intellectual disability were assessed at 6-month intervals. Data were analysed using qualitative and statistical analyses.Four groups of individuals emerged from the analysis: (1) confirmed suspicion, (2) no suspicion, (3) questionable suspicion and (4) early suspicion. Significant differences were found between groups 1 and 2. The WDTIM could be administered to 90%-100% of all participants exhibiting mild-to-moderate intellectual disability and to 50% with severe intellectual disability .The WDTIM was shown to have good applicability to people with mild-to-moderate intellectual disability and to be appropriate for detecting cognitive changes. Using the two instruments in combination achieved greater accuracy in reinforcing a dementia suspicion than did using the DSQIID alone.
Pub.: 24 Mar '17, Pinned: 25 Aug '17
Abstract: The work represents the potent catalytic activity of silver nanoparticles synthesized from Cicer arietinum (chickpea) leaf extract (CAL-AgNPs). Here, silver nano-catalysts were used against the anthropogenic pollutants mainly involving nitro-amines and azo dyes. These pollutants are extremely harmful to our environment and causes severe health issues. The CAL-AgNPs have the potential to degrade harmful toxins and their by-products, thereby decreasing the pollutants from the environment. The green synthesis of nano-catalyst includes a simple, cost effective and eco-friendly method using the leaf extract from the plant. A systematic study was conducted, including synthesis, optimization and characterization of the silver particles. The AgNPs were further assessed through DLS and TEM for size and morphological evaluation. The obtained particles have shown spherical morphology with the size range of 88.8nm. Further, FTIR were performed for compositional and functional group analysis of the particles. The antibacterial efficiency was also evaluated against E. coli and P. aeruginosa. For their catalytic evaluation, CAL-AgNPs were assessed for 4-nitrophenol, methylene blue and congo red. The results obtained through catalytic evaluation suggested that the CAL-AgNPs could be helpful to surmount the environmental pollution in a very effective manner.
Pub.: 02 Aug '17, Pinned: 25 Aug '17