Workshop on Un/Semi-supervised learning and Data Mining (14-15 October 2019)
This focused workshop serves the presentation of research performed in WP3 (Unsupervised and semi-supervised learning) and the coordination of further activities. It is hosted by AXA at their Paris premises: 61, rue Mstislav Rostropovich, Paris 75017.
Tentative Schedule of Workshop
13.00. Welcome and Intro (I. Emiris and AXA)
13.15. Invited Talk (TBD)
14.15. Coffee break
14.30. Dangerous Curve Extraction (M. Sammarco, AXA)
15.30. TBD (AXA)
16.30. Fast, safe and efficient routing (N. Zygouras, NKUA)
17.30. End of 1st day
Evening. Social Dinner.
09.30. Learning Language for Shape Differentiation (P. Achlioptas, Stanford)
10.30. Coffee Break
11.00. Road segmentation, clustering & abnormality detection (I. Chamodrakas, NKUA)
11.30 Clustering of one-dimensional objects (E. Christoforou, NKUA )
12.30. Data-driven methods for financial modeling (A. Chalkis, NKUA)
13.30. Lunch Break
14.30. Conclusions, Outlook and Plans (ALL)
15.30 End of the workshop
Midterm Meeting (25-28 September 2018)
It serves for reporting and making sure execution of the Project follows the expected plans. It is hosted by 3DI at their London HQ: 1 Berkeley Street, Mayfair, London. On this occasion, the Midterm Review occurs (Thursday 27/9 afternoon to Friday 28/9 noon).
The meeting includes a General Software Workshop (25-27/9), which is a means to coordinate innovation activities based on software development, to review software standards that must be followed by developers in view of integrating their methods to libraries, and to discuss topics related to scaling up our methods for addressing big data. The Workshop includes talks by our members, our USA partners, and by invited speaker Prof. Anne Driemel (TU Eindhoven).
Schedule of Workshop
Tu. 25/9: Arrival
W. 26/9, 10.00-12.30: Informal discussions
W. 26/9, 14.00-16.00: Consortium talks
W. 26/9, 16.00-18.00: Board meeting
Th. 27/9, morning: Invited speakers
9.00 – 10.00 Panagiotis Achlioptas, Stanford University (Remotely)
Learning Fine-Graned Distinctions of 3D Objects from Referential Language
Human world knowledge is highly structured and flexible. When people see an object, they represent it not as a pixel array but as a coherent object consisting of semantic parts. Similarly, when people refer to an object, they provide descriptions that are not merely true but relevant in the current context. In this work, we harness human behavior in a challenging, grounded reference game task to learn fine-grained correspondences between language and the contextually relevant geometric properties of structured 3D objects. We demonstrate high-performing neural listeners and speakers and introduce a novel dataset containing utterances referring to objects from ShapeNet across a wide diversity of contexts. Using targeted lesions of visual and linguistic input, we show that our systems learn to depend on part words and associate them with the corresponding geometric elements. Moreover, we find that a context-sensitive neural speaker that plans utterances according to how an imagined listener would interpret their words, in context, produces high quality shape descriptions.
10.15 – 11.15 Anne Driemel (TU Eindhoven, http://www.win.tue.nl/~adrieme)
Searching under the Frechet distance
Similarity search is a fundamental data analysis task. When the input data consists of trajectories or time series, the standard Euclidean distance is unable to capture the similarity properly, unless the data is heavily normalized, preprocessed, or observed only in a feature space. In order to make the similarity search meaningful on the raw data, it is desirable to support more flexible distance measures such as the Frechet distance or Dynamic Time Warping (DTW). Unfortunately, it is an open problem how to build efficient data structures that support similarity search under the Frechet distance. In this talk I will present some recent advances in this area, in particular, I will present a volume argument showing a lower bound on the space-query-time tradeoff and practical approaches based on LSH that might serve as a remedy. Time permitting, I will also present some recent results on the related problem of clustering curves under the Frechet distance.
11.30-12.30 Consortium talks
Venue for the Workshop on Wednesday
Schedule of Midterm Review
Thursday 27 Sep.
14:00-14:10 Round the table Introduction (All)
14:10-14:40 Intro: Status, Financial/administrative issues, ethics, secondments, risks, deviations, dissemination, networking, Deliverables, Tasks, Milestones (Emiris)
14:40-15:00 WP1: Rigorous methods and software (I. Emiris, WP Leader; I. Psarros, secondee)
15:00-15:15 Coffee Break
15:15-15:35 WP2: Retrieval and Shape analysis (S. Rejal, WP Leader; E. Anagnostopoulos, secondee)
15:35-15:55 WP3: Unsupervised and semi-supervised learning (M. Detyniecki, WP Leader; I. Chamodrakas, E. Christophorou, secondees)
15:55-16:15 WP4a. Training: Transfer of Knowledge and Networking, secondments, transfer of knowledge, career opportunities (S. Rejal, WP Leader)
16.15-16.35 WP4b. Dissemination and Communication (S. Rejal, WP Leader)
16:35-16:50 Coffee Break
16:50-17.10 WP5: Project Management (I. Emiris, WP Leader)
17.10-17:30 WP6: Ethics Requirement (I. Emiris, WP Leader)
Friday 28 Sep.
09:30-10:30 Meeting between secondees and the REA Project Officer (Secondees)
In parallel: Meeting between site leaders and Innovation expert, discussion on pre-identified innovation aspects (see questionnaire)
10:30-11.00 Coffee Break
11.00-12.00 Open Discussion (All)
12.00-12:30 Summary of Agreed action points (Emiris)
12:30-13.00 Recommendations Overview (Project Officer)
13.00-14.00 Project Lunch
Venue for the Midterm Review
Citadines Holborn-Covent Garden London, located on 94-99 High Holborn, London WC1V 6LF near the Holborn Tube station (contact phone number 0207 395 8800). Conference rooms are located on the ground floor past the reception. Please note that the workshop will be held on Thursday 27th in the “Sydney” conference room and Friday 28th in the “Berlin” conference room.
Kickoff General Workshop (6-8 June 2017)
It serves for academic and industrial members to present the expertise they bring into LAMBDA as well as open problems in which they are working, thus finalising the planned work. Moreover, it serves for organising technology transfer, benchmarking, and joint software development. There shall be an invited speaker.
The meeting will be held on Wednesday, 7th June 2017 and Thursday 8th June 2017, 09.00-17.00. Since participants are expected to arrive on Tuesday, June 6th, a social event may be planned on Tuesday evening. Sessions on Wednesday and Thursday shall be held in the A56 conference room of the Department of Informatics & Telecommunications, at the Campus of National & Kapodistrian University of Athens at Ilisia.
Confirmed participants from all beneficiaries include: M. Detyniecki, I. Emiris, D. Gunopulos, I. Hamodrakas, S. Rejal, X. Renard. Partner participant: Panos Achlioptas (Stanford U.). The ethics adviser has also confirmed participation. Invited speakers: Prof. Joachim Giesen, and Dr. Soeren Laue (Univ. Jena, Germany).
Wed. 9.00. Welcome and Intro (Emiris). Team of NKUA (Emiris, Gunopulos).
10.00. COFFEE BREAK
10.30. Team of 3DI (Rejal).
11.30. Team of AXA (Detyniecki, Renard).
12.30. LUNCH BREAK
14.30. Proximity problems in high dimensions (Psarros, NKUA)
15.30. Web-scale clustering (Anagnostopoulos, NKUA)
16.30. ORGANISATIONAL DISCUSSION
20.00. SOCIAL DINNER at “Mpousoulas”
Thu. 9.00. Invited talk: Scaling Up Generic Optimization for Data Analytics (Giesen and Laue).
10.00. COFFEE BREAK
10.30. Supervisory board meeting
11.30. Predictive Learning for 3D Point Clouds (Achlioptas).
12.30. LUNCH BREAK
14.30. A General Framework for First Story Detection Utilizing Entities and their Relations (Panagiotou, NKUA)
15.30. ChEsS: Cost-Effective Scheduling across multiple heterogeneous mapreduce clusters (Zachilas, Kalogeraki, AUEB)
16.30. CONCLUSIONS AND OUTLOOK
Lunch will be provided on Wednesday and Thursday, as well as a workshop dinner on Wednesday evening. A social event may be organised on Tuesday evening.
Please arrange for your own hotel accommodation (unless you are invited). Suggested hotels in the area (in order of increasing price): Delice Hotel & Apartments (website). Ilisia Best-Western. Divani-Caravel Hotel, on Vassileos Alexandrou Ave. Athens Hilton Hotel.
Directions from Athens International Airport: see also transport from the airport: The best is to use Metro Line 3, and get off at “Evangelismos” stop. Timetable: 05:30-24:00; frequency: 30′. Taxis: their stand is at the Arrivals Level; fee: 40 to 50 Euro. With the Express Airport Bus X95 (to Syntagma sq.): get off at the Hilton Bus Stop; it runs on a 24-hour basis.
Directions to Dept. Informatics & Telecoms from the hotels. There’s 2 bus lines for which you need a valid metro ticket or a bus ticket, to be bought at a kiosk before entering the bus. Bus 250 goes in the campus, drops you off in front of the department (Bus Stop: 2nd PANEPISTIMIOPOLIS). Alternatively use “224”, get off at stop “10th KESSARIANIS”, on “Ethnikis Antistaseos” Ave, turn left and walk 100m. Here is a Google map to the Dept of Informatics from “Evangelismos” metro stop (or the hotels). See also the School of Science.
Satellite event: Summer school “Techniques in Random Discrete Structures“, University of Athens, Greece, May 22-27, 2017.
Focused Workshop on 3D shape analysis & Focused Workshop on Insurance data mining
The two Focused Workshops occur for each of the main application domains, organised respectively by the leaders of WP2 and WP3 at their sites. All teams involved in WP2 or WP3 shall participate in the respective meeting. The one on 3D shape analysis involves NKUA, 3DI, OSU, SU, while the one on insurance data mining involves NKUA, AXA, UC. The exact date is not fixed, but we give tentative dates in September and October 2019, respectively.
Final General Workshop
It is an important means of dissemination and communication, hence it shall be collocated with a major conference, to be determined in the course of the Project. Candidate meetings, with whom we have experience as participants, program committee members and organisers are the following international conference: KDD (ACM Knowledge Discovery and Data Mining), NIPS (Neural Information Processing Systems, a leading conference in Machine Learning), ACM SIGRAPH (Graphics), SoCG (Symposium on Computational Geometry). All of them are leaders in their respective field. This increases our visibility. Tentatively in December 2020.