Brazil

IWSSIP 2020

Evaluation of Output Representations in Neural Network-based Trajectory Predictions Systems

output representations

lstms

trajectory prediction

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-23 is the Thirty-Seventh AAAI Conference on Artificial Intelligence. The theme of this conference is to create collaborative bridges within and beyond AI. Like previous AAAI conferences, AAAI-23 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and two new activities: a Bridge Program and a Lab Program. Many of these activities are tailored to the theme of bridges and all are selected according to the highest standards, with additional programs for students and young researchers. 
AAAI is providing you with a conference planner, which you can use to help organize your itinerary of activities. This includes talks to attend in person, talks to attend remotely, breaks with colleagues and your site seeing activities. To access this conference planner, please go to [https://aaai-2023.takemobi.io](https://aaai-2023.takemobi.io).

In order to access this site, you need to register. If you haven't already, please register [here](https://aaai.org/Conferences/AAAI-23/registration/).


AAAI 2023

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines.

technical paper

The 27th International Conference on Systems, Signals and Image Processing.

Brazilian Electricity Regulatory Agency (ANEEL) defines as non-technical losses the problems that happen during the energy consumption measurement process. A feasible alternative to reduce those losses is the energy consumption reading process performed by customers, called self-reading. This process might be executed across digital platforms such as mobile applications and the customers would register and send consumption information. In this context, the consumption reading using voice is a simple way for the public that has lower affinity for technology to perform self-reading. Therefore, this work proposes an end-to-end speech recognition method applied to the energy consumption measurement that uses a neural network architecture based on Recurrent Neural Network Transducer (RNN-T). The proposed method uses a modified architecture called Residual Recurrent Neural Network Transducer (RRNN-T) which contains a Residual Skip that make its behavior similar to a network shallower than RNN-T by skipping some layers in the training process. That strategy saves computational cost of the inference process and disk space. A spelled sequence of digits is the input of the proposed method that recognizes those digits and outputs the electrical energy consumption reading measured in kilowatts per hour. The dataset used in this work contains 111,224 samples of spelled sequences divided into 70\% for training and 30\% for test. The proposed method obtains 0.05 word error rate (WER) and inference time less than 3 seconds on mobile devices.

Self-Reading by Speech: An End-to-end Speech Recognition Method Applied to the Energy Consumption Measurement on Mobile Devices

Intelligent transport support systems have had a major impact on people’s urban mobility. In large urban centers, transportation services still need ways to optimize vehicle supply in certain areas, according to the demand in each of them. This work proposes a stream learning methodology for taxi demand prediction, that tackles concept drift detection on a real taxi data stream.

A Methodology for Taxi Demand Prediction Using Stream Learning

Smart cities should be built on harmonious relationships with the environment, with rational measures to use environmental resources in many ways. Resources essential to human life such as water and energy must have special attention. With the advance of technologies use, human being becomes one of the main promoters of information and solutions for smart cities. Stream machine learning is an approach where data becomes available sequentially, called data stream, which adapts the prediction model with fixed window-sized data stream. The objective of this work is to analyze the effectiveness of buildings energy consumption using stream learning. To achieve this goal, MOA tool was used to evaluate the performance of the selected stream learning algorithms on an an open database. Batch algorithms were also used as testbed, to analyze the performance of stream algorithms related to them

Energy Consumption Prediction using Data Stream Learning for Commercial Buildings

Monitoring and using city devices to improve the life-quality of citizens is currently a challenge.
Such devices, as traffic lights and cameras, have been providing data to assist decisions to improve the functionality of cities.
Verifying and certifying the systems used for critical decisions is one of the challenges that still require some effort.
Reo takes advantage of a formal logic-based framework which makes it possible to reason, guarantee requirements, absence of described failures and provide certified software (i.e.\ proved-to-be-correct regarding the specification).
This work is tailored to show how to use Reo to model Smart Cities software interaction compiling to an automated logic model able to guarantee the required constraints in a certified system.

Modelling and Certifying Smart Cities in Reo Circuits

Wi-Fi networks have attracted a lot of attention due to its large deployment. In this work, we evaluate the performance of Multi-label machine learning methods to build classification models capable of providing wireless network access points (AP) occupancy predictions throughout the day. The APs association dataset used in our experiment was built using data from SCIFI network of Fluminense Federal University for a period of 6 months.

An Experimental Analysis for Detecting Wi-Fi Network Associations Using Multi-label Learning

One of the most critical issues in Smart Cities (SC) is how to use the energy system properly. The capacity to provide more detailed consumption information and more accurate consumer feedback is an essential step towards energy sustainability. On the end-user side, a system that supports the measurement of several points of consumption and provides visual information can help consumers reduce energy waste. On the other side, electric utilities can use a real-time measurement system to infer the energy consumption behavior, to find metering inconsistencies, to optimize the system, or to maintain the reliability of the supply system based on data information. This work presents a demand-side management Framework for smart cities which aims to provide integration between consumers and energy distribution utilities. %with prototype and proof of concept,
This Framework can help the electricity demand control, presenting real-time consumption data and presenting relevant data on their consumption points to consumers. A prototype was developed as a proof of concept.

Demand-Side Management Framework for Smart Cities

A digital watermark is an information or signal inserted into a digital content (image, audio, video, document, software) that can be used to determine its ownership, showing who created or sold it, or to ensure its integrity. Typically, the same watermark is added consistently to mark movie collections. Consequently, media with a visible watermark will be eventually susceptible to the necessity of watermark remission, operation that could be an issue. This paper presents two new and very
fast algorithms for watermark detection and removal. The here considered watermarks could be blended, overlaid or added in a continuous form to the digital video on remission treatment. The first of such algorithms use a template matching approach to pre-process the blended watermarks frame-to-frame in the Hue-Saturation-Value (HSV) color space, and then improve the detection using a post-processing step. The second algorithm presents two steps as well. Initially, it estimates the structure of a translucent watermark observing the constancy of its gradient magnitude in the area corresponding to the tenth part of each video. Secondly, it uses a sliding window to detect the watermarked in the frames. The results show that watermark detection by both algorithms has high sensitivity. Moreover, they are very fast when compared with other methods especially those based on deep learning

Watermark Detection and Clearance in Video Using Simple Signal and Image Processing Techniques

The development of techniques to automatic set up parameters in image processing methods based on variational approaches is cumbersome because the model sensitivity to parameters values is not known in general. In this paper, we address this issue through a data-driven inverse problem. Specifically, our methodology receives pairs (input image(s), desired result(s)) and seeks for the near optimum parameter vector through an inverse problem based on minimization scheme. The methodology is not restricted to a particular functional and it does not require large annotated data sets as input. Besides, methods based on partial differential equations (PDEs) can be also accommodated in our approach. We validate the methodology for calibrating parameters using as test-bed a variation of Mumford-Shah method. The obtained solutions using the parameters found in this paper are compared with literature results in order to show the efficiency of our technique.


Parameterizing Variational Methods Through Data-Driven Inverse Problems for Image Processing Applications.

Breast cancer is one of the most severe types ofcancer and responsible for about 15% of all women death related to cancer worldwide. Mammography is the most reliable and recommended approach to help early detection of breast cancer, since this technique can detect the disease in its asymptomatic phase. Different Computer-Aided Diagnosis (CAD) systems have been used to aid in the diagnosis serving as second opinion to the physicians. These systems extract features from masses found in mammograms, which compose inputs to the classifiers used in CAD systems. The main contribution of this study is the evaluation of different polygonal representations on the phase of feature extraction. In this context, two different polygonal models were tested with different parameters to represent boundaries of masses. Artificial neural networks, support vector machines and k-nearest neighbors were used to discriminate masses as benign and malignant achieving accuracy of 85%, 84% and 85%, respectively.

Evaluating the impact of polygonal representations on mass classification

In this presentation we evaluate an approach of transfer learning to automatic semantic segmentation of lungs in a pediatric chest x-ray image based on
knowledge acquired in other datasets. The results are promising and
reach 0.917 of Dice using U-Net in a chest x-ray dataset without lung annotation

Premium content

Evaluation of Output Representations in Neural Network-based Trajectory Predictions Systems

Downloads

Next from IWSSIP 2020

Self-Reading by Speech: An End-to-end Speech Recognition Method Applied to the Energy Consumption Measurement on Mobile Devices

Similar lecture

Causal Intervention for Human Trajectory Prediction with Cross Attention Mechanism

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES