We propose a data-driven scene flow estimation algorithm exploiting the observation that many 3D scenes can be explained by a collection of agents moving as rigid bodies. At the core of our method lies a deep architecture able to reason at the object-level by considering 3D scene flow in conjunction with other 3D tasks. This object level abstraction, enables us to relax the requirement for dense scene flow supervision with simpler binary background segmentation mask and ego-motion annotations. Our mild supervision requirements make our method well suited for recently released massive data collections for autonomous driving, which do not contain dense scene flow annotations. As output, Figure 1: Our network takes two successive frames as input our model provides low-level cues like pointwise flow (a), and outputs a set of transformation parameters for each and higher-level cues such as holistic scene understanding segmented rigid agent (c) which are used to recover perpoint at the level of rigid objects. We further propose a test-time rigid scene flow. After applying the predicted flow to optimization refining the predicted rigid scene flow. We the first point cloud, the two frames are aligned (b, d).

This paper addresses the problem of media retrieval using a multimodal query (a query which combines visual input with additional semantic information in natural language feedback). We propose a SynthTriplet GAN framework which resolves this task by expanding the multimodal query with a synthetically generated image that captures semantic information from both image and text input. We introduce a novel triplet mining method that uses a synthetic image as an anchor to directly optimize for embedding distances of generated and target images. We demonstrate that apart from the added value of retrieval illustration with synthetic image with the focus on customization and user feedback, the proposed method greatly surpasses other multimodal generation methods and achieves state of the art results in the multimodal retrieval task. We also show that in contrast to other retrieval methods, our method provides explainable embeddings.

Device identification is one way to secure a network of IoT devices, whereby devices identified as suspicious can subsequently be isolated from a network. We introduce a novel fingerprinting method, IoT-DevID, for device identification that uses machine learning to model the behaviour of IoT devices based on network packets. Our method uses an enhanced combination of features from previous work and includes an approach for dealing with unbalanced device data via data augmentation. We further demonstrate how to enhance device identification via a group-wise data aggregation. We provide a comparative evaluation of our method against two recent identification methods using three public IoT datasets which together contain data from over 100 devices. Through our evaluation we demonstrate improved performance over previous results with F1-scores above 99%, with considerable improvement gained from data aggregation.

Autonomous navigation is a key skill for assistive and service robots. To be successful, robots have to minimise the disruption caused to humans while moving. This implies predicting how people will move and complying with social conventions. Avoiding disrupting personal spaces, people's paths and interactions are examples of these social conventions. This paper leverages Graph Neural Networks to model robot disruption considering the movement of the humans and the robot so that the model built can be used by path planning algorithms. Along with the model, this paper presents an evolution of the dataset SocNav1 which considers the movement of the robot and the humans, and an updated scenario-to-graph transformation which is tested using different Graph Neural Network blocks. The model trained achieves close-to-human performance in the dataset. In addition to its accuracy, the main advantage of the approach is its scalability in terms of the number of social factors that can be considered in comparison with handcrafted models.

The application of remaining useful life (RUL) prediction has taken great importance in terms of energy optimization, cost-effectiveness, and risk mitigation. The existing RUL prediction algorithms mostly constitute deep learning frameworks. In this paper, we implement LSTM and GRU models and compare the obtained results with a proposed genetically trained neural network. The current models solely depend on Adam and SGD for optimization and learning. Although the models have worked well with these optimizers, even little uncertainties in prognostics prediction can result in huge losses. We hope to improve the consistency of the predictions by adding another layer of optimization using Genetic Algorithms. The hyper-parameters - learning rate and batch size are optimized beyond manual capacity. These models and the proposed architecture are tested on the NASA Turbofan Jet Engine dataset. The optimized architecture can predict the given hyper-parameters autonomously and provide superior results.

Market by order (MBO) data - a detailed feed of individual trade instructions for a given stock on an exchange - is arguably one of the most granular sources of microstructure information. While limit order books (LOBs) are implicitly derived from it, MBO data is largely neglected by current academic literature which focuses primarily on LOB modelling. In this paper, we demonstrate the utility of MBO data for forecasting high-frequency price movements, providing an orthogonal source of information to LOB snapshots. We provide the first predictive analysis on MBO data by carefully introducing the data structure and presenting a specific normalisation scheme to consider level information in order books and to allow model training with multiple instruments. Through forecasting experiments using deep neural networks, we show that while MBO-driven and LOB-driven models individually provide similar performance, ensembles of the two can lead to improvements in forecasting accuracy -- indicating that MBO data is additive to LOB-based features.

Generalization is a central problem in Machine Learning. Indeed most prediction methods require careful calibration of hyperparameters usually carried out on a hold-out \textit{validation} dataset to achieve generalization. The main goal of this paper is to introduce a novel approach to achieve generalization without any data splitting, which is based on a new risk measure which directly quantifies a model's tendency to overfit. To fully understand the intuition and advantages of this new approach, we illustrate it in the simple linear regression model ($Y=X\beta+\xi$) where we develop a new criterion. We highlight how this criterion is a good proxy for the true generalization risk. Next, we derive different procedures which tackle several structures simultaneously (correlation, sparsity,...). Noticeably, these procedures \textbf{concomitantly} train the model and calibrate the hyperparameters. In addition, these procedures can be implemented via classical gradient descent methods when the criterion is differentiable w.r.t. the hyperparameters. Our numerical experiments reveal that our procedures are computationally feasible and compare favorably to the popular approach (Ridge, LASSO and Elastic-Net combined with grid-search cross-validation) in term of generalization. They also outperform the baseline on two additional tasks: estimation and support recovery of $\beta$. Moreover, our procedures do not require any expertise for the calibration of the initial parameters which remain the same for all the datasets we experimented on.

Deep learning techniques achieve high accuracy in computer vision tasks. However, their accuracy suffers considerably when they face a domain change, i.e., as soon as they are used in a domain that differs from their training domain. For example, a road sign recognition model trained to recognize road signs in Germany performs poorly in countries with different road sign standards like China. We propose ConTraKG, a neuro-symbolic approach that enables cross-domain transfer learning based on prior knowledge about the domain or context. A knowledge graph serves as a medium for encoding such prior knowledge, which is then transformed into a dense vector representation via embedding methods. Using a five-phase training pipeline, we train the deep neural network to adjust its visual embedding space according to the domain-invariant embedding space of the knowledge graph based on a contrastive loss function. This allows the neural network to incorporate training data from different target domains that are already represented in the knowledge graph. We conduct a series of empirical evaluations to determine the accuracy of our approach. The results show that ConTraKG is significantly more accurate than the conventional approach for dealing with domain changes. In a transfer learning setup, where the network is trained on both domains, ConTraKG achieves 21% higher accuracy when tested on the source domain and 15% when tested on the target domain compared to the standard approach. Moreover, with only 10% of the target data for training, it achieves the same accuracy as the cross-entropy-based model trained on the full target data.

As the worldwide population gets increasingly aged, in-home tele-medicine and mobile-health solutions represent promising services to promote active and independent aging and to contribute to a paradigm shift towards a patient-centric healthcare. In this work, we present ACTA (Advanced Cognitive Training for Aging), a prototypal mobile-health solution to provide advanced cognitive training for senior citizens with mild cognitive impairments, We disclose here the conceptualization of ACTA as the integration of two promising rehabilitation strategies: the "Nudge theory", from the cognitive domain, and the neurofeedback, from the neuroscience domain. Moreover, in ACTA we exploit the most advanced machine learning techniques to deliver customized and fully adaptive support to the elderly, while training in an ecological environment. ACTA represents the next-step beyond SENIOR, an earlier mobile-health project for cognitive training based on Nudge theory, currently ongoing in Lombardy Region. Beyond SENIOR, ACTA represents a highly-usable, accessible, low-cost, new-generation mobile-health solution to promote independent aging and effective motor-cognitive training support, while empowering the elderly in their own aging. As the worldwide population gets increasingly aged, tele-medicine and mobile-health solutions are becoming key services to promote active and independent aging, and to contribute to a paradigm shift towards a patient-centric healthcare. In many countries, especially Italy, Portugal in Europe, and Japan in Asia, average population age is rapidly increasing and projections indicate 79% of it will be over 60 by 2050, according to the 2017 report of the United Nations [1]. Mild cognitive impairment (MCI) is rapidly becoming one of the most common clinical manifestations affecting the elderly. It is characterized by deterioration of memory, attention, and cognitive function that is beyond what is expected based on age and educational level. MCI does not interfere significantly with individuals' daily activities. It can act as a transitional level towards dementia with a range of conversion of 10%-15% per year.

Since their inception, learning techniques under the Reservoir Computing paradigm have shown a great modeling capability for recurrent systems without the computing overheads required for other approaches. Among them, different flavors of echo state networks have attracted many stares through time, mainly due to the simplicity and computational efficiency of their learning algorithm. However, these advantages do not compensate for the fact that echo state networks remain as black-box models whose decisions cannot be easily explained to the general audience. This work addresses this issue by conducting an explainability study of Echo State Networks when applied to learning tasks with time series, image and video data. Specifically, the study proposes three different techniques capable of eliciting understandable information about the knowledge grasped by these recurrent models, namely, potential memory, temporal patterns and pixel absence effect. Potential memory addresses questions related to the effect of the reservoir size in the capability of the model to store temporal information, whereas temporal patterns unveils the recurrent relationships captured by the model over time. Finally, pixel absence effect attempts at evaluating the effect of the absence of a given pixel when the echo state network model is used for image and video classification. We showcase the benefits of our proposed suite of techniques over three different domains of applicability: time series modeling, image and, for the first time in the related literature, video classification. Our results reveal that the proposed techniques not only allow for a informed understanding of the way these models work, but also serve as diagnostic tools capable of detecting issues inherited from data (e.g. presence of hidden bias).