Annonces du cours Geometric Data Analysis

Youtube series of computational geometry concept videos

Sat, 02 Sep 2023 17:17:33 +0300

By Carola Wenk & Maarten Löffler: Each video considers a geometric concept and discusses its history, complexity, and other interesting properties. The videos will be released weekly on Maarten's Youtube channel https://www.youtube.com/@maartenloeffler. This season consists of nine videos. The first video is on the Closest & Farthest Pair https://www.youtube.com/watch?v=kx4RM5owuwY&t=1110s

School on machine learning for shapes

Fri, 29 Jan 2021 15:34:06 +0300

Next week the first doctoral school of GRAPES will take place remotely. The topic is «machine learning for shapes». Ηave a look at this webpage http://grapes-network.eu/event/doctoral-school-i-midterm-meeting/ and let me know if you would like to attend -- the school will not be widely open so we will send you the connection information.

-γ-

Ανάθεση εργασίας για τελική αξιολόγηση.

Thu, 07 Jan 2021 14:42:42 +0300

Kαλημερα και καλη χρονια..

σε συνεργασια με τον κ. Εμιρη, στο πλαισιο της αξιολογησης του μαθηματος προτεινουμε την βιβλιογραφικη αναδρομη σε 3 θεματα (φαινονται παρακατω) με πυρηνα τις δημοσιευσεις που προτεινονται σε καθε ενα απο αυτα.

Η αξιολογηση σας θα γινει (σε ημερομηνια που θα ανακοινωθει εγκαιρα) με βαση μια 30 λεπτη (+ ερωτησεις) παρουσιαση.

Οι 5 φοιτητες/τριες που παρακολουθουν το μαθημα με στοχο την βαθμολογηση τους να χωριστειτε μετα απο συνεννοηση μεταξυ σας σε 3 ομαδες (πχ 2+2+1, η ομαδα με 1 ατομο θα εχει προφανως λιγοτερο φορτο) - και να μας ενημερωσετε μεχρι 12 Ιανουαριου με ενα μηνυμα εμαιλ προς τον κ. Εμιρη με αντιγραφο σε μενα.

Ευχες,

Μ. Βαζιργιαννης
===================

1. Graph convolutional Networks

- The first approach for a graph neural network model came from
Scarcelli 2009 [1].
- The most impactful work after Scarcelli was from Kipf & Welling 2016
[2] , who introduced a matrix-form spectral based neural network for
graphs called GCN, which incorporated the graph topology as matrix
multiplication.
- Transforming the GCN into label propagation schemes was a next crucial
step, such as GraphSAGE from Hamilton 2017 [3]. This work highlighted a
first connection between the spectral-based GCN and the spatial-based
GNNs, which ( the latter ) are described via locally aggregational rules.

[1]
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.1015.7227&rep=rep1&type=pdf
[2]
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.1015.7227&rep=rep1&type=pdf
1

[3] https://arxiv.org/abs/1706.02216

2. Graph attention networks and expressiveness

- Most of the works until 2017 described isotropic graph learning
architectures, that treated the neighborhoods of each node equally. As a
complement to that, Velickovic 2017 proposed Graph Attention Networks
[4], where the definition of graph attentions was built. Â
- Since 2018, there was an increasing interest in measuring and
extending the expressivity of graph neural networks. Xu et al 2018/2019
[5] presented one of the first efforts of defining the expressive power
with respect to the equivalence with WL-isomorphism test.
- On parallel with the comparison with the WL-isomorphism test, efforts
have been made to understand the meaning of depth in GNNs and why most
deep GNN models fail to perform equally to shallow ones, such as the
findings of Oono 2019/2020 [6].

[4] https://arxiv.org/abs/1710.10903
[5] https://arxiv.org/abs/1810.00826
[6] https://arxiv.org/pdf/1905.10947.pdf

3. Graph based methods for NLP

Graphs have proven very effective for NLP tasks. The following papers
present the original approach [10] and aspects such as keyword
extraction[7] and also GNN based approaches [9].

[7]Main core retention on graph-of-words for single-document keyword
extraction, F Rousseau, M Vazirgiannis European Conference on
Information Retrieval, 382-393
[9] Message passing attention networks for document understanding, G
Nikolentzos, A Tixier, M Vazirgiannis. Proceedings of the AAAI
Conference on Artificial Intelligence 34 (05), 8544-8551
[10] Graph-of-word and TW-IDF: new approach to ad hoc IR, F Rousseau, M
Vazirgiannis, Proceedings of the 22nd ACM CIKM, 140, 2013

P. Achlioptas talks on 8/1 at 11.15

Sun, 05 Jan 2020 20:24:56 +0300

Καλησπερα και Καλη χρονια!

Panos Achlioptas (Stanford) will give a Lab seminar on Wednesday 8/1, room A56 at 11.15, on:

State-of-Art methods for self-supervised learning in 2D: All ideas are directly applicable on 3D as well.

tomorrow (Thursd 22/10) presentations

Wed, 21 Oct 2020 11:31:59 +0300

Projects have been assigned on webpage "Documents".
We expect 10-15' presentations by each of you tomorrow on how you started your project
Slides are allowed, otherwise you may use the blackboard if needed.

Σεμινάριο Παρασκευή 20 Δεκεμβρίου, στις 12.15, αίθουσα α56

Wed, 11 Dec 2019 23:40:47 +0300

TITLE: Autonomous Driving: Simulation and Navigation

Dinesh Manocha

Department of Computer Science and Electrical & Computer Engineering
University of Maryland at College Park
http://gamma.umd.edu/ad

Abstract:
Autonomous driving has been an active area of research and development
over the last decade. Despite considerable progress, there are many open
challenges including automated driving in dense and urban scenes. In
this talk, we give an overview of our recent work on simulation and
navigation technologies for autonomous vehicles. We present a novel
simulator, AutonoVi-Sim, that uses recent developments in physics-based
simulation, robot motion planning, game engines, and behavior modeling.
We describe novel methods for interactive simulation of multiple
vehicles with unique steering or acceleration limits taking into account
vehicle dynamics constraints. We present techniques for navigation with
non-vehicle traffic participants such as cyclists and pedestrians. Our
approach facilitates data analysis, allowing for capturing video from
the vehicle's perspective, exporting sensor data such as relative
positions of other traffic participants, camera data for a specific
sensor, and detection and classification results. We highlight its
performance in traffic and driving scenarios. We also present novel
multi-agent simulation algorithms using reciprocal velocity obstacles
that can model the behavior and trajectories of different traffic agents
in dense scenarios, including cars, buses, bicycles and pedestrians. We
also present novel methods for extracting trajectories from videos and
use them for behavior modeling and safe navigation. These techniques are
based on spectral analysis and demonstrated on urban datasets
corresponding to ArgoVerse and TRAF.

Brief Biography:
Dinesh Manocha is the Paul Chrisman Iribe Chair in Computer Science &
Electrical and Computer Engineering at the University of Maryland
College Park. He is also the Phi Delta Theta/Matthew Mason Distinguished
Professor Emeritus of Computer Science at the University of North
Carolina - Chapel Hill. He has won many awards, including Alfred P.
Sloan Research Fellow, the NSF Career Award, the ONR Young Investigator
Award, and the Hettleman Prize for scholarly achievement. His research
interests include multi-agent simulation, virtual environments,
physically-based modeling, and robotics. He has published more than 520
papers and supervised more than 36 PhD dissertations. He is an inventor
of 10 patents, several of which have been licensed to industry. His work
has been covered by the New York Times, NPR, Boston Globe, Washington
Post, ZDNet, as well as DARPA Legacy Press Release. He is a Fellow of
AAAI, AAAS, ACM, and IEEE, member of ACM SIGGRAPH Academy, and Pioneer
of Solid Modeling Association. He received the Distinguished Alumni
Award from IIT Delhi the Distinguished Career in Computer Science Award
from Washington Academy of Sciences. He was a co-founder of Impulsonic,
a developer of physics-based audio simulation technologies, which was
acquired by Valve Inc in November 2016. See http://www.cs.umd.edu/~dm

Υπολογιστική Αλγεβρα

Thu, 07 Nov 2019 17:02:02 +0300

Επισημαίνω πως υπάρχει το ιδιαίτερα ενδιαφέρον (και πολύ γεωμετρικό) Μεταπτυχιακό μάθημα κάθε Παρασκευή 12-3μμ, ενταγμένο στο ΠΜΣ Τμήματος και το ΑΛΜΑ: https://eclass.uoa.gr/courses/D231/

Today: Course starts at 3pm sharp

Thu, 11 Feb 2021 14:15:24 +0300

Today we have presentations by the students, starting at 3.00 pm (sharp) and ending by 5pm. We'll devote about 20' per student. We'll send a ZOOM link before the class begins.

Yiannis

ps. Apologies for the late notice.

PhD defence on Proximity problems for high-dimensional data

Sat, 08 Jun 2019 01:07:53 +0300

Ο Υποψήφιος διδάκτορας Γιάννης Ψαρρός, παρουσιάζει την διδακτορική του διατριβή με τίτλο
Proximity problems for high-dimensional data

την Δευτέρα 10/6 στις 16.00, στην αιθουσα Α56 του Τμήματος Πληροφορικής. Ακουλουθει η περίληψη.
Συναδελφικά,

-γ-

Abstract:
With the recent increase of availability of complex datasets, the need
for analyzing and handling high-dimensional descriptors has been
increased. Likewise, there is a surge of interest into data structures
for trajectory processing, motivated by the increasing availability and
quality of trajectory data. In this thesis, we investigate proximity
problems for high-dimensional vectors and polygonal curves. In
particular, the largest part of this thesis is devoted to the approximate
nearest neighbor problem and the approximate near neighbor
problem. We also study a problem of computing good representatives for a
pointset, and we study the VC dimension for range spaces defined by
polygonal curves.

We introduce a new definition of “low-quality” embeddings for metric
spaces. It requires that, for some query point q, there exists an
approximate nearest neighbor among the pre-images of the k > 1
approximate nearest neighbors in the target space. Focusing on Euclidean spaces, we employ
random projections à la Johnson-Lindenstrauss in order to reduce the original problem to one in a space of dimension inversely proportional

to k. This leads to simple data structures which are space-efficient and
also support sublinear queries. We also propose a dimension reduction
by means of a near neighbor-preserving embedding for pointsets with low
intrinsic dimension, under the Manhattan distance. Finally, we extend
recent results for the approximate closest pair problem to the problem
of computing approximate r-nets, by reducing the problem to multi-point
evaluation of polynomials.

For polygonal curves under the discrete Fréchet and Dynamic Time Warping
distance functions, we offer the first data structures and query
algorithms for the approximate nearest neighbor problem with arbitrarily
good approximation factor, at the expense of increasing space usage and
preprocessing time over existing methods. In the case of the discrete
Fréchet distance, we also present an improved result which is sensitive
to the complexity of the query. Finally, we bound the VC-dimension for
range spaces defined by polygonal curves: several implications follow
due to already known sampling results.

SIFT dataset for the programming assignment

Tue, 07 May 2019 17:49:41 +0300

A dataset has been uploaded for the programming assignment in the Documents section of the e-class page of the course. The dataset is derived from SIFT and contains objects in 128 dimensions. The zip file also contains a query set. A good value for the w parameter of the Euclidean LSH is between 300 and 400 for the specific dataset.

Hypercube random projection experimental results for the SIFT dataset:

dim = 11, p = 20: MAF= 2.7264345

dim = 11, p = 30: MAF = 1.7987736

dim= 8, p = 20: MAF = 2.4184265

dim= 8, p = 50: MAF = 1.158912

data analytics job

Mon, 04 Mar 2019 11:44:25 +0300

Στην ΕΛΒΑΛ (Ελληνική Βιομηχ. Αλουμινιου), ενδιαφέρονται να προσλάβουν κάποιον για data analytics.
Αν καποιος ενδιαφερεται ας επικοινωνήσει μαζι μου,
-γ-