My task is to organise the eClub Summer Camp. Select an interesting project for each of the students. Find the best mentor for the selected project. Make sure all have the right tools. Help everybody to get productive and deliver the best results. I also interact with our sponsors creating interface between students and partners experts. I need to understand our partners business needs to make them happy securing continuous cooperqaation and support.
I am focusing on research of human machine interaction devices in automotive environment and its application in improvement of driver's experience and reduction the number of dangerous situations in traffic.
Finding correct answers for your questions is time consuming task. You have to search through many web pages and check, whether your found answer is true. My goal is to create mobile client for question answering system YodaQA, which allows you to ask questions in natural language and get answer in the same way.
Long Hoang Nguyen
Question Answering is a computer science discipline combining information retrieval and natural language processing, which aims to automatically answer questions posed by humans. It has recently made waves when IBM’s Watson won at Jeopardy. I work on an open source question answering system called YodaQA as a general worker bee. My current goal is to improve the expressiveness of our web interface by adding additional information to our answers, such as the source text. Ultimately, we want the users to see the exact reasons why YodaQA chose each answer.
Our task is to design and develop smart infrastructure for management of various IOT devices. Main goals are minimal effort from end user during management, compatibility with 3rd party devices and also provide web API which would allow development of applications using data from our IOT infrastructure.
IOT subject is very popular these days. Even though home automatization exists in some way, there is still a lot to improve. The task over the Summer Camp is to develop working IOT home automatization system. Main goal is automatic sensor detection and integration into the system. The system should provide secure sensor connection and managment with focus on easy of use.
My project focuses on finding new ways of powering wireless sensors. Today's state of the art in terms of powering electronic devices are Lithium based batteries. However, the number of smart sensors in homes is still rising and finding a new ways of powering those is a big issue today. Another field of use are sensors for dangerous environments, implantable sensors etc. I do research in three fields - TEG (thermoelectric), RF (electromagnetic waves) and light/laser harvesting. The result of the research will be a proof-of-the-concept battery-free Bluetooth enabled sensor.
Internet of Things is sometimes being called "the next big thing in IT", as the possibilities in this field are virtually endless. However, even if the technology itself is great, the idea won't sell when it is too complicated to use. For the interconnection of new devices, sensors and other appliances to a "smart home", we are developing (among other things) mobile application designed to handle connecting with a minimal effort from the user.
Question answering is very challenging problem. It consists of natural language processing and information retrieval. The generated set of answers is scored based on computed probabilities. The correct answer should be on the top positions. I work in the YodaQA team where my goal is to improve the performance of the question answering system. To do so, we are trying various classifiers for answer scoring and new answer sources.
Query rewrite / expansion is a technique of reformulation a original query to improve the performance of the search system. The query can be rewritten by many approaches, such as: query words synonyms, words stemming, spelling correction, product synonyms, etc. I will focus on using this technique in the E-commerce search system.
Search engines (SE) are being used as a powerful and indispensable tool for hundreds of millions of people for searching information on the internet. The amount of information amassed by search engines attracts users with malicious behavior targeting to e.g. exploit web site vulnerabilities or collect email addresses. Other malicious 'users' are attracted by the search engines importance for advertising and are trying to e.g. deplete competitions advertising budget or promote their webpages SE result page's rank or affect the query autocompletion system. Detecting atypical users is a non-trivial task due to the following reasons: The size of the daily traffic of the SE The distinct targets and nature of the atypical users The lack/absence of annotated data The fact that SE has to be available to all users 24/7 In recent years, multiple scientific papers were published on this topic. My bachelor thesis covers the most significant ones. Most of the works use methods of unsupervised learning, as there is no publicly available labeled dataset. The approach used in my bachelor thesis and also in other works, is based on user session modelling. This approach seems natural, as we are trying to identify anomalous users. This time we would like to take an approach of suspicious URL detection, which could later serve as a first step in an atypical user detection pipeline. We would like to build a system that will use statistical tests and time series prediction to detect URLs whose traffic differs significantly from their history.
The goal of this project is to develop wireless controller designated to remote control of workplace illumination. The controller will not serve only as a simple on/off switch, but it will also provide advanced functions such as feedback regulation of illumination based on ambient light, manual intensity-based regulation or proximity-based regulation. To ensure robustness, the controller will operate the light only via wireless HUB and thus it will be potentially capable of regulating various kinds of lights. Main benefits of such solution for potential customers would be: no need of complicated installation of wiring into walls, portability of the controller, automated regulation providing constant illumination over time and energy savings thanks to human detection.
The goal of the project is to create an automated intelligent assistant (bot) through which the user will be able to create an appointment in selected business. The application should connect Google Calendar service with Facebook Messenger and the bot program itself. The user will communicate with the bot application via Facebook Messenger, and will book an appointment through an intelligent dialogue system. The bot application will retrieve all necessary data from Google Calendar. The application will also mark the agreed meeting in the calendar. This project should simplify appointment booking by making the interaction with one of the employees no longer necessary.
As a part of conversion of the YodaQA system to answering questions in Czech, my task is its deployment on a new server using Docker and structuring the code of the system. I will deploy each of YodaQA's modules, test them and replace them with improved versions if needed.
PubChem is a publicly available database of molecules, containing information about broad range of their chemical properties. To date, it describes over 80 million chemical compounds, each of them represented by line notation called SMILES. While some attributes like molecular weight can be computed directly from this formula, there is a large number of properties that have to be obtained experimentally. The goal of this project is to predict these values using recurrent neural networks, which proved to be ideal for use with string based input, and will benefit from extent of the PubChem database.
Knowledge graphs representing real-world entities and their relationships as vertices and edges, are useful for many applications. A simple example is answering factoid questions application (e.g., “Where was the president of France born?”). My project is automatically extending an incomplete knowledge graph (Wikidata) with new true facts, yielding a larger, more useful knowledge graph. This can be done both by reasoning over the graph itself (e.g., the nationality of people born in cities in Germany is most likely German) or with the help of additional unstructured or semi-structured data (e.g., Wikipedia articles, news).
We have multiple projects at EClub which would benefit from having a tool that could provide us with the intent of a sentence/question, for example QA ( "Who wrote Ender's game?"), bots ("What is the weather in Prague?") or even IOT ("Set AC to 25 degrees"). I would like to use my previous experience with the task of Sentence Pair Similarity to create such tool.
Automated bots are new way for interacting with technologies. Bots can help users to perform routine tasks easily through known interfaces like messengers. In this project we want to focus on a conversational commerce. Assistance during the selection and purchase of a product is one of the tasks, which can be solved by bot. The bot will help the user to specify product parameters, it will explain product details, advantages, and lead the user to a selection and finally purchase. We have decidet to design a bot helping in choosing the right smartphone.
Recent application trends in IT are heading towards using bots. Supposing a bot used in a specific field like recommending phones, this bot parses a request from a user, discovers the intent of the request and tries to reply. And since human requests have usually syntactic order as subject-predicate-object, RDF databases seem as logical choice for storing information for such a domain and preserving its semantic meaning. Challenge accepted in this project is to create a complete RDF database for some domain that could be then exploited by a bot using SPARQL semantic queries.