Greece

Human activity recognition and analysis has always been one of the most active areas of pattern recognition and machine intelligence, with applications in various fields, including but not limited to exertion games, surveillance, sports analytics and healthcare. Especially in Human-Robot Interaction, human activity understanding plays a crucial role as household robotic assistants are a trend of the near future. However, state-of-the-art infrastructures that can support complex machine intelligence tasks are not always available, and may not be for the average consumer, as robotic hardware is expensive. In this paper we propose a novel action sequence encoding scheme which efficiently transforms spatio-temporal action sequences into compact representations, using Mahalanobis distance-based shape features and the Radon transform. This representation can be used as input for a lightweight convolutional neural network. Experiments show that the proposed pipeline, when based on state-of-the-art human pose estimation techniques, can provide a robust end-to-end online action recognition scheme, deployable on hardware lacking extreme computing capabilities.

EANN 2020

A compact sequence encoding scheme for online human activity recognition in HRI applications

radon transform

action recognition

human-robot interaction

deep neural networks

ICPR 2020 is the premier international conference on Pattern Recognition, gathering leading researchers from around the world to present their latest achievements in Pattern Analysis and Recognition, applied Artificial Intelligence and Machine Learning. The topic of the year is " Putting Artificial intelligence to work on Patterns". 
For the first time, ICPR has employed a two-round review process similar to journal submissions, with new papers allowed to be submitted in either the first or second round and papers submitted in the first round and not accepted allowed to be revised and re-submitted for second round review. 

In this edition of the **ICPR 2020** Conference, there are 1407 Oral/Poster presentations organized in five Tracks: 
- Artificial Intelligence, Machine Learning for Pattern Analysis; 
- Biometrics, Human Analysis and Behavior Understanding; Computer Vision, Robotics and Intelligent Systems; 
- Document and Media Analysis; and 
- Image and Signal Processing. 

The **ICPR 2020** Main Conference program also includes 7 Keynotes from prominent world-renowned scientists, 8 Tutorials, 28 Demos, 8 Challenges, 2 Panels, as well as an Industry/Sponsor Exhibition. In addition, 41 Workshops on a range of important current and emerging research topics complete the ICPR 2020 program. 


ICPR 2020

technical paper

21st edition of EANN (International Conference on Engineering Applications of Neural Networks) should be held in Halkidiki, Greece. Due to current Covid-19 situation the conference will be held virtually and Underline will host a repository of video talks and preserve them for the future. Videos will be released on Underline one day after the conference. 

This paper presents a self-sufficient solar-powered real-time Deep Learning (DL) based system that runs inference 100% on solar energy and which is composed of an Nvidia Jetson TX2 board and a dual-axis solar tracker based on the cast-shadow principle. In order to have a higher energy being generated by the solar tracker as well as a lower energy consumption by the real-time DL-based system, we have: a) updated our solar tracker’s panel with a higher number of polycrystalline photovoltaic (PV) cells and connected it to a chain of two inverters, one accumulator and one solar charge controller; b) implemented a motion detection method that triggers the inference process only when there is substantial movement in webcam frame. Experimental results show that our solar tracker generates sufficient and constant solar energy for all the 4 DL models (VGG-19, Incep-tionV3, ResNet-50 and MobileNetV2) that are running in real-time on the Nvidia Jetson TX2 platform and which requires more than 5 times less energy when compared to a laptop having a Nvidia GTX 1060 GPU, proving that real-time DL-based systems can be powered by solar trackers without the need of traditional power plugs or need to pay for electricity bills.

Efficient Implementation of a Self-Sufficient Solar-Powered Real-Time Deep Learning-Based System

The main aim of this paper is to study the stability problem for neutral-type Hopﬁeld neural networks possessing multiple time delays in the states of the neurons and multiple neutral delays in time derivative of states of the neurons. By using a suitable Lyapunov functional, a novel suﬃcient stability condition is obtained for global asymptotic stability of neutral-type neural networks with multiple delays. The derived stability criterion can be expressed in terms of the parameters of the neural network model which totally relies on some simple relationships established between the network parameters and it is completely independent of time delays and neutral delays. Hence, this new global asymptotic stability condition can be easiliy tested and veriﬁed by using some algebraic mathematical properties. We will also make a comparision between the result of this paper and previously published corresponding results. This comparision will indicate the advanteges of our proposed stability condition over the previously reported stability conditions. Since obtaining stability conditions for neutral type neural networks with multiple delays is a diﬃcult task to achieve due to the insuﬃcient mathematical methods and techniques, the result given in this paper can be considered an important and alternative result for this class of neutral type neural systems.

Stability Analysis of Neutral-Type Hopﬁeld Neural Networks with Multiple Delays

Radar sensors, unlike lidars or cameras, can measure objects' instantaneous velocity and composition. However, the ability to process and retrieve spatial information from radar point clouds has not yet been fully established. In this work, we propose a key technique for improving the performance of standard machine learning point-wise processing methods on radar point clouds. We show that a network can learn to extract object-related signatures for every point using automotive radar measurements. In addition, we propose RadarPCNN, a novel architecture for performing semantic segmentation, specifically designed for radar point clouds. RadarPCNN uses PointNet++, aided with mean shift as feature extractor module and an attention mechanism to fuse information from different neighborhood levels. We show that our model outperforms state-of-the-art solutions on our dataset.

Leveraging Radar Features to Improve Point Clouds Segmentation with Neural Networks

Unbalanced electrical loads on feeders of power electrical systems can cause serious problems, including power losses, significantly lower power quality, damaging of electrical equipment, and tripping of protective devices. Nevertheless, the problem of balancing such systems ---which essentially is equivalent to the problem of three-way integer partitioning--- has proven to be NP-complete. Against this background, in this article, an algorithm based on the powerful, declarative framework of Answer Set Programming (ASP) is provided, that efficiently attacks practical instances of the phase-balancing problem. To the best of our knowledge, this is the first attempt of approaching this significant engineering problem by means of the ASP paradigm. The whole study indicates that the examined problem is of great interest from an algorithmic viewpoint, as well as an engineering application that highlights ASP's modelling methodology.

An ASP-Based Approach for Phase Balancing in Power Electrical Systems

Public Key Encryption with Keyword Search (PEKS) allows users to search encrypted files by a specific keyword without compromising
the original data security. Almost all current PEKS schemes enable users to search exact keyword only instead of imprecise keyword (such as
“latest”, “biggest”, etc.). Therefore, if the keyword is fuzzy, these PEKS schemes will be terminated and then report errors. Besides, some PEKS
schemes are not secure mainly because they are vulnerable to Off-line Keyword Guessing Attack (OKGA). This research paper incorporates
with Mamdani Fuzzy Inference System to PEKS for supporting Fuzzy Keyword Search. Secondly, the proposed scheme is proved to be semantic
secure under the random oracle models so that it is able to resist OKGA. In addition, the new scheme allows users to search multiple keywords and therefore, it could be applied to the general public networks.

Fuzzy Logic Application to Searchable Cryptography (PEKS)

Common types of artificial neural networks have been well known to suffer from the presence of outlying measurements (outliers) in the data. However, there are only a few available robust alternatives for training common form of neural networks. In this work, we investigate robust fitting of multilayer perceptrons, i.e. alternative approaches to the most common type of feedforward neural networks. Particularly, we consider robust neural networks based on the robust loss function of the least trimmed squares, for which we express formulas for 
derivatives of the loss functions. Some formulas, which are however incorrect, have been already available. Further, we consider a very recently proposed multilayer perceptron based on the loss function of the least weighted squares, which appears a promising highly robust approach. We also derive the derivatives of the loss functions, which are to the best of our knowledge a novel contribution of this paper. The derivatives may find applications in implementations of the robust neural networks, if a (gradient-based) backpropagation algorithm is used.

Robust multilayer perceptrons: Robust loss functions and their derivatives

A fully-connected Deep Neural Network (DNN) architecture, to be referred to as λ-DNN, used to predict 2D/3D scalar fields is presented. In aerodynamics, the λ-DNN is firstly trained on fields computed using a Computational Fluid Dynamics (CFD) software. Then, it can be incorporated into engineering processes in various ways. One possibility is to use them in optimization problems solved by stochastic population-based methods, in which the λ-DNN may act as the surrogate evaluation model, replacing calls to the CFD tool. Another possibility is in multi-disciplinary problems, to replicate the numerical solver for any of the disciplines. This small list of possible usages is not exhaustive and, of course, different usages can be combined. The input to each DNN contains information to identify the geometrical shape and case-related data, nodal coordinates and, in multi-disciplinary problems, interfacing data connecting solvers for different disciplines on adjacent domains. In this presentation, the λ-DNN is firstly used in the aerodynamic shape optimization of a wing using evolutionary algorithms, in which it replicates the CFD solver. Then, it is used in a conjugate heat transfer problem dealing with a solid domain in contact with a flow within a duct. In this problem, the λ-DNN acts as a surrogate to the solver of the heat conduction equation on the solid domain, by interfacing with a CFD solver of the fluid domain.

λ-DNNs and their Implementation in Aerodynamic and Conjugate Heat Transfer Optimization

Recent advances in optical sensor technologies and those of Geoinformatics can now support very large scale high definition multispectral and panchromatic images. This capability allows the use of remote sensing for the observation of complex ecosystems on earth. It supports sustaining biodiversity, precision agriculture, the management of land, crops, and parasites. It also supports advanced quantitative studies of biophysical and biogeochemical cycles in coastal or inland waters. One of the key applications that will allow the development of new types of decision support systems, is that of precise and effective scene classification. This will offer considerable advantages to business, science, and engineering. This work proposes a novel and very effective system based on geographic object-based scene classification in remote sensing images. More specifically the proposal includes a deep systems architecture of the Residual Neural Network (ResNet) type. This, omitting some layers in the early stages of training, achieves an effective simplification of the network by eliminating the Vanishing Gradient Problem (VGP) which troubles other deep learning architectures. Furthermore, the proposed system uses the softmax activation instead of the sigmoid function in the last level and the fully residual network was trained using the novel AdaBound algorithm that employs dynamic bounds on their learning rates. This achieves a smooth transition to stochastic gradient descent, tackling the noise dispersed points of misclassification with great precision. This is something that other spectral classiﬁcation methods cannot handle. The proposed system has been tested, very successfully, in scene identification from remote sensing images. This confirms that it could be further used in advanced level processes for Large-Scale Geospatial Data Analysis such as cross-border classification, the recognition, and monitoring of certain patterns and multisensor data fusion.

Large-Scale Geospatial Data Analysis: Geographic Object-based Scene Classification in Remote Sensing Images by GIS and Deep Residual Learning

Critical Care Units (CCU) in a hospital treat the severely sick patients that need constant monitoring and close medical attention. Feeding patients, enteral feeding in particular, is a critical and continu- ous process. Monitoring patients, managing their feeding and referring to a dietician is a key factor in CCUs. Screening patients for referral to a dietician in a CCU is an error-prone and complicated task. One of the main challenges in this regard is that the data needed to screen patients is scattered among many different variables and textual forms. The number of patients being treated in the CCU is also a significant problem since it becomes difficult for the staff to keep track of the needs of all patients. Therefore, an automated screening tool can support effectively the feeding process and contribute considerably towards improving the quality and consistency of patient care. In this paper we present early stages of a project that aims at using machine learning techniques to help CCU consultants to automatically screen patients for dietician referral.

Automated Screening of Patients for Dietician Referral

Received a bachelor's (2016) and master's degree (2018) from ITMO University, specialization: software engineering. Currently, I am a research programmer at the ITMO University machine learning laboratory and DeepPavlov laboratory at MIPT.

Downloads

Next from EANN 2020

Efficient Implementation of a Self-Sufficient Solar-Powered Real-Time Deep Learning-Based System

Similar lecture

Attention-Oriented Action Recognition for Real-Time Human-Robot Interaction

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from EANN 2020

Efficient Implementation of a Self-Sufficient Solar-Powered Real-Time Deep Learning-Based System

Similar lecture

Attention-Oriented Action Recognition for Real-Time Human-Robot Interaction

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads