Publications
Results
- Showing results for:
- Reset all filters
Search results
-
Conference paperPardo F, Levdik V, Kormushev P, 2020,
Scaling all-goals updates in reinforcement learning using convolutional neural networks
, 34th AAAI Conference on Artificial Intelligence (AAAI 2020), Publisher: Association for the Advancement of Artificial Intelligence, Pages: 5355-5362, ISSN: 2374-3468Being able to reach any desired location in the environmentcan be a valuable asset for an agent. Learning a policy to nav-igate between all pairs of states individually is often not fea-sible. Anall-goals updatingalgorithm uses each transitionto learn Q-values towards all goals simultaneously and off-policy. However the expensive numerous updates in parallellimited the approach to small tabular cases so far. To tacklethis problem we propose to use convolutional network archi-tectures to generate Q-values and updates for a large numberof goals at once. We demonstrate the accuracy and generaliza-tion qualities of the proposed method on randomly generatedmazes and Sokoban puzzles. In the case of on-screen goalcoordinates the resulting mapping from frames todistance-mapsdirectly informs the agent about which places are reach-able and in how many steps. As an example of applicationwe show that replacing the random actions inε-greedy ex-ploration by several actions towards feasible goals generatesbetter exploratory trajectories on Montezuma’s Revenge andSuper Mario All-Stars games.
-
BookDeisenroth MP, Faisal AA, Ong CS, 2020,
Mathematics for Machine Learning
, Publisher: Cambridge University Press, ISBN: 9781108455145 -
Conference paperSaputra RP, Rakicevic N, Kormushev P, 2020,
Sim-to-real learning for casualty detection from ground projected point cloud data
, 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2019), Publisher: IEEEThis paper addresses the problem of human body detection-particularly a human body lying on the ground (a.k.a. casualty)-using point cloud data. This ability to detect a casualty is one of the most important features of mobile rescue robots, in order for them to be able to operate autonomously. We propose a deep-learning-based casualty detection method using a deep convolutional neural network (CNN). This network is trained to be able to detect a casualty using a point-cloud data input. In the method we propose, the point cloud input is pre-processed to generate a depth image-like ground-projected heightmap. This heightmap is generated based on the projected distance of each point onto the detected ground plane within the point cloud data. The generated heightmap-in image form-is then used as an input for the CNN to detect a human body lying on the ground. To train the neural network, we propose a novel sim-to-real approach, in which the network model is trained using synthetic data obtained in simulation and then tested on real sensor data. To make the model transferable to real data implementations, during the training we adopt specific data augmentation strategies with the synthetic training data. The experimental results show that data augmentation introduced during the training process is essential for improving the performance of the trained model on real data. More specifically, the results demonstrate that the data augmentations on raw point-cloud data have contributed to a considerable improvement of the trained model performance.
-
Journal articleStimberg M, Goodman D, Nowotny T, 2020,
Brian2GeNN: accelerating spiking neural network simulations with graphics hardware
, Scientific Reports, Vol: 10, Pages: 1-12, ISSN: 2045-2322“Brian” is a popular Python-based simulator for spiking neural networks, commonly used in computational neuroscience. GeNNis a C++-based meta-compiler for accelerating spiking neural network simulations using consumer or high performance gradegraphics processing units (GPUs). Here we introduce a new software package, Brian2GeNN, that connects the two systems sothat users can make use of GeNN GPU acceleration when developing their models in Brian, without requiring any technicalknowledge about GPUs, C++ or GeNN. The new Brian2GeNN software uses a pipeline of code generation to translate Brianscripts into C++ code that can be used as input to GeNN, and subsequently can be run on suitable NVIDIA GPU accelerators.From the user’s perspective, the entire pipeline is invoked by adding two simple lines to their Brian scripts. We have shown thatusing Brian2GeNN, two non-trivial models from the literature can run tens to hundreds of times faster than on CPU.
-
Conference paperJohns E, Liu S, Davison A, 2020,
End-to-end multi-task learning with attention
, The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019, Publisher: IEEEWe propose a novel multi-task learning architecture, which allows learning of task-specific feature-level attention. Our design, the Multi-Task Attention Network (MTAN), consists of a single shared network containing a global feature pool, together with a soft-attention module for each task. These modules allow for learning of task-specific features from the global features, whilst simultaneously allowing for features to be shared across different tasks. The architecture can be trained end-to-end and can be built upon any feed-forward neural network, is simple to implement, and is parameter efficient. We evaluate our approach on a variety of datasets, across both image-to-image predictions and image classification tasks. We show that our architecture is state-of-the-art in multi-task learning compared to existing methods, and is also less sensitive to various weighting schemes in the multi-task loss function. Code is available at https://github.com/lorenmt/mtan.
-
Book chapterCocarascu O, Toni F, 2020,
Deploying Machine Learning Classifiers for Argumentative Relations “in the Wild”
, Argumentation Library, Pages: 269-285Argument Mining (AM) aims at automatically identifying arguments and components of arguments in text, as well as at determining the relations between these arguments, on various annotated corpora using machine learning techniques (Lippi & Torroni, 2016).
-
Conference paperNadler P, Arcucci R, Guo Y, 2020,
An Econophysical Analysis of the Blockchain Ecosystem
, Pages: 27-42, ISSN: 2198-7246We propose a novel modelling approach for the cryptocurrency ecosystem. We model on-chain and off-chain interactions as econophysical systems and employ methods from physical sciences to conduct interpretation of latent parameters describing the cryptocurrency ecosystem as well as to generate predictions. We work with an extracted dataset from the Ethereum blockchain which we combine with off-chain data from exchanges. This allows us to study a large part of the transaction flows related to the cryptocurrency ecosystem. From this aggregate system view we deduct that movements on the blockchain and price and trading action on exchanges are interrelated. The relationship is one directional: On-chain token flows towards exchanges have little effect on prices and trading volume, but changes in price and volume affect the flow of tokens towards the exchange.
-
Journal articleZambelli M, Cully A, Demiris Y, 2020,
Multimodal representation models for prediction and control from partial information
, Robotics and Autonomous Systems, Vol: 123, ISSN: 0921-8890Similar to humans, robots benefit from interacting with their environment through a number of different sensor modalities, such as vision, touch, sound. However, learning from different sensor modalities is difficult, because the learning model must be able to handle diverse types of signals, and learn a coherent representation even when parts of the sensor inputs are missing. In this paper, a multimodal variational autoencoder is proposed to enable an iCub humanoid robot to learn representations of its sensorimotor capabilities from different sensor modalities. The proposed model is able to (1) reconstruct missing sensory modalities, (2) predict the sensorimotor state of self and the visual trajectories of other agents actions, and (3) control the agent to imitate an observed visual trajectory. Also, the proposed multimodal variational autoencoder can capture the kinematic redundancy of the robot motion through the learned probability distribution. Training multimodal models is not trivial due to the combinatorial complexity given by the possibility of missing modalities. We propose a strategy to train multimodal models, which successfully achieves improved performance of different reconstruction models. Finally, extensive experiments have been carried out using an iCub humanoid robot, showing high performance in multiple reconstruction, prediction and imitation tasks.
-
Conference paperJha R, Belardinelli F, Toni F, 2020,
Formal verification of debates in argumentation theory.
, Publisher: ACM, Pages: 940-947 -
Conference paperArcucci R, Casas CQ, Xiao D, et al., 2020,
A Domain Decomposition Reduced Order Model with Data Assimilation (DD-RODA)
, Conference on Parallel Computing - Technology Trends (ParCo), Publisher: IOS PRESS, Pages: 189-198, ISSN: 0927-5452 -
Conference paperNadler P, Arcucci R, Guo Y-K, 2020,
A Scalable Approach to Econometric Inference
, Conference on Parallel Computing - Technology Trends (ParCo), Publisher: IOS PRESS, Pages: 59-68, ISSN: 0927-5452- Author Web Link
- Cite
- Citations: 1
-
Conference paperArcucci R, Mottet L, Casas CAQ, et al., 2020,
Adaptive Domain Decomposition for Effective Data Assimilation
, 25th International Conference on Parallel and Distributed Computing (Euro-Par), Publisher: SPRINGER INTERNATIONAL PUBLISHING AG, Pages: 583-595, ISSN: 0302-9743 -
Book chapterArcucci R, Moutiq L, Guo Y-K, 2020,
Neural Assimilation
, Editors: Krzhizhanovskaya, Zavodszky, Lees, Dongarra, Sloot, Brissos, Teixeira, Publisher: SPRINGER INTERNATIONAL PUBLISHING AG, Pages: 155-168, ISBN: 978-3-030-50432-8- Author Web Link
- Cite
- Citations: 2
-
Conference paperCocarascu O, Cabrio E, Villata S, et al., 2020,
Dataset Independent Baselines for Relation Prediction in Argument Mining.
, Publisher: IOS Press, Pages: 45-52 -
Conference paperLiu S, Davison A, Johns E, 2019,
Self-supervised generalisation with meta auxiliary learning
, 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Publisher: Neural Information Processing Systems Foundation, Inc.Learning with auxiliary tasks can improve the ability of a primary task to generalise.However, this comes at the cost of manually labelling auxiliary data. We propose anew method which automatically learns appropriate labels for an auxiliary task,such that any supervised learning task can be improved without requiring access toany further data. The approach is to train two neural networks: a label-generationnetwork to predict the auxiliary labels, and a multi-task network to train theprimary task alongside the auxiliary task. The loss for the label-generation networkincorporates the loss of the multi-task network, and so this interaction between thetwo networks can be seen as a form of meta learning with a double gradient. Weshow that our proposed method, Meta AuXiliary Learning (MAXL), outperformssingle-task learning on 7 image datasets, without requiring any additional data.We also show that MAXL outperforms several other baselines for generatingauxiliary labels, and is even competitive when compared with human-definedauxiliary labels. The self-supervised nature of our method leads to a promisingnew direction towards automated generalisation. Source code can be found athttps://github.com/lorenmt/maxl.
-
Journal articleRakicevic N, Kormushev P, 2019,
Active learning via informed search in movement parameter space for efficient robot task learning and transfer
, Autonomous Robots, Vol: 43, Pages: 1917-1935, ISSN: 0929-5593Learning complex physical tasks via trial-and-error is still challenging for high-degree-of-freedom robots. Greatest challenges are devising a suitable objective function that defines the task, and the high sample complexity of learning the task. We propose a novel active learning framework, consisting of decoupled task model and exploration components, which does not require an objective function. The task model is specific to a task and maps the parameter space, defining a trial, to the trial outcome space. The exploration component enables efficient search in the trial-parameter space to generate the subsequent most informative trials, by simultaneously exploiting all the information gained from previous trials and reducing the task model’s overall uncertainty. We analyse the performance of our framework in a simulation environment and further validate it on a challenging bimanual-robot puck-passing task. Results show that the robot successfully acquires the necessary skills after only 100 trials without any prior information about the task or target positions. Decoupling the framework’s components also enables efficient skill transfer to new environments which is validated experimentally.
-
Conference paperNadler P, Arcucci R, Guo YK, 2019,
Data assimilation for parameter estimation in economic modelling
, Pages: 649-656We propose a data assimilation approach for latent parameter estimation in economic models. We describe a dynamic model of an economic system with latent state variables describing the relationship of economic entities over time as well as a stochastic volatility component. We show and discuss the model's relationship with data assimilation and how it is derived. We apply it to conduct a multivariate analysis of the cryptocurrency ecosystem. Combining these approaches opens a new dimension of analysis to economic modelling. Economics, Multivariate Analysis, Dynamical System, Bitcoin, Data Assimilation.
-
Conference paperLim EM, Molina Solana M, Pain C, et al., 2019,
Hybrid data assimilation: An ensemble-variational approach
, Pages: 633-640Data Assimilation (DA) is a technique used to quantify and manage uncertainty in numerical models by incorporating observations into the model. Variational Data Assimilation (VarDA) accomplishes this by minimising a cost function which weighs the errors in both the numerical results and the observations. However, large-scale domains pose issues with the optimisation and execution of the DA model. In this paper, ensemble methods are explored as a means of sampling the background error at a reduced rank to condition the problem. The impact of ensemble size on the error is evaluated and benchmarked against other preconditioning methods explored in previous work such as using truncated singular value decomposition (TSVD). Localisation is also investigated as a form of reducing the long-range spurious errors in the background error covariance matrix. Both the mean squared error (MSE) and execution time are used as measure of performance. Experimental results for a 3D case for pollutant dispersion within an urban environment are presented with promise for future work using dynamic ensembles and 4D state vectors.
-
Journal articleAristodemou E, Arcucci R, Mottet L, et al., 2019,
Enhancing CFD-LES air pollution prediction accuracy using data assimilation
, Building and Environment, Vol: 165, ISSN: 0007-3628It is recognised worldwide that air pollution is the cause of premature deaths daily, thus necessitating the development of more reliable and accurate numerical tools. The present study implements a three dimensional Variational (3DVar) data assimilation (DA) approach to reduce the discrepancy between predicted pollution concentrations based on Computational Fluid Dynamics (CFD) with the ones measured in a wind tunnel experiment. The methodology is implemented on a wind tunnel test case which represents a localised neighbourhood environment. The improved accuracy of the CFD simulation using DA is discussed in terms of absolute error, mean squared error and scatter plots for the pollution concentration. It is shown that the difference between CFD results and wind tunnel data, computed by the mean squared error, can be reduced by up to three order of magnitudes when using DA. This reduction in error is preserved in the CFD results and its benefit can be seen through several time steps after re-running the CFD simulation. Subsequently an optimal sensors positioning is proposed. There is a trade-off between the accuracy and the number of sensors. It was found that the accuracy was improved when placing/considering the sensors which were near the pollution source or in regions where pollution concentrations were high. This demonstrated that only 14% of the wind tunnel data was needed, reducing the mean squared error by one order of magnitude.
-
Journal articlePeach R, Yaliraki S, Lefevre D, et al., 2019,
Data-driven unsupervised clustering of online learner behaviour 
, npj Science of Learning, Vol: 4, ISSN: 2056-7936The widespread adoption of online courses opens opportunities for analysing learner behaviour and optimising web-based learning adapted to observed usage. Here we introduce a mathematical framework for the analysis of time series of online learner engagement, which allows the identification of clusters of learners with similar online temporal behaviour directly from the raw data without prescribing a priori subjective reference behaviours. The method uses a dynamic time warping kernel to create a pairwise similarity between time series of learner actions, and combines it with an unsupervised multiscale graph clustering algorithm to identify groups of learners with similar temporal behaviour. To showcase our approach, we analyse task completion data from a cohort of learners taking an online post-graduate degree at Imperial Business School. Our analysis reveals clusters of learners with statistically distinct patterns of engagement, from distributed to massed learning, with different levels of regularity, adherence to pre-planned course structure and task completion. The approach also reveals outlier learners with highly sporadic behaviour. A posteriori comparison against student performance shows that, whereas high performing learners are spread across clusters with diverse temporal engagement, low performers are located significantly in the massed learning cluster, and our unsupervised clustering identifies low performers more accurately than common machine learning classification methods trained on temporal statistics of the data. Finally, we test the applicability of the method by analysing two additional datasets: a different cohort of the same course, and time series of different format from another university.
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.