Research on using Normalizing Flows to simulate detector response of particle jet deposits. Using the generative capabilities of recent Normalizing Flow models like MAF and BNAF to model conplex jet distributions encountered when simulating detector deposits.

We explore compositional reasoning on the CATER dataset by learning action embeddings and by using object centric representations.

Abstract: We present a survey of ways in which existing scientific knowledge are included when constructing models with neural networks. The inclusion of domain-knowledge is of special interest not just to constructing scientific assistants, but also, many other areas that involve understanding data using human-machine collaboration. In many such instances, machine-based model construction may benefit significantly from being provided with human-knowledge of the domain encoded in a sufficiently precise form. This paper examines the inclusion of domain-knowledge by means of changes to: the input, the loss-function, and the architecture of deep networks. The categorisation is for ease of exposition: in practice we expect a combination of such changes will be employed. In each category, we describe techniques that have been shown to yield significant changes in the performance of deep neural networks.

ECMWF (Europe's largest meteorological research institute) acts a data vendor to several clients, providing massive amounts of meteorological data to them. We work on building intelligent Anomaly Detection systems capable of monitoring the log files produced by these services for sudden disruptions and failures.

This work was supported by a grant of £5,000 from ECMWF.

The Bongard problems were introduced by Mikhail Moiseevich Bongard in 1967, in his classic pattern recognition book. We use the DeepProbLog framework to model solutions to these problems by evaluating different hypotheses and their respective likelihoods.

Schizophrenia is a mental disorder whith symptoms including hallucinations and episodes of psychosis. We develop a Deep Learning Pipeline for automated Schizophrenia detection using abnormalies in brain-wave patterns captured through EEG Data.

Spike-Timing Dependent Plasticity, or STDP is the proposed theory which aims to relate temporal spike differences to changes in synaptic weights between participating neurons. We explore how well STDP works on a model of the Basal ganglia, using the Izhikevich neuron, using SpineCreator to model the underlying networks.

Developed a pipeline for Deep Time Series Anomaly Detection on Server Log files. Integrated several State of the Art Anomaly Detection and Forecasting algorithms with a Realtime plotting framework into a python package. This work was funded by ECMWF's open source program - ESoWC.

Developed a Deep Learning pipeline for Emotion recognition and classification using speech data, on the MELD Dataset. Classified emotions across various emotions : [Disgust, Fear, Neutral, ...] across a highly unbalanced data sample. Used Mel-frequency cepstral coefficients (MFCCs) to form speech representations.

Integrated deep text and image processing models to build a Multimodal Sentiment Analysis system that classified emotions on Internet Memes across different categories. This work was done on the Sem-Eval dataset.

A modular implementations of Conway's Game of Life in Python with common patterns like Still lifes, Oscillators, and Spaceships.

Developed a Siamese Neural Net that performed few shot signature verification. Taking a few sample signatures of a person, the model predicted whether the input 'query' signature fas forged or authentic.

Worked with genomic data from different geographical locations, plotting it after dimensionality reduction. Demonstrated relations between geographic origin and DNA structure by generating different plots and identifying aggregations.

Developed a Generative Adversarial Network (GAN) to generate new instances of the CIFAR Dataset.

Generated word embeddings using GloVe, and the Large Movie Review Dataset. Visualized the obtained embeddings on a 2D graph using Principal Component Analysis (PCA), and checked the obtained embeddings for semantic coherence.