HUN-REN SZTAKI, 1111. Kende utca 13-17., Grand Hall
The presentation slides are available here and videos of the presentations are available here.
MILAB in cooperation with the Society of Image Processors and Shape Recognition will hold a professional day on 19 June in the basement of the HUN-REN SZTAKI, Kende utca 13-17.
Participation is subject to registration, a registration link will be sent to those who have subscribed to MILAB events, sign up:
https://limesurvey.sztaki.hu/index.php?r=survey/index&sid=976865
The preliminary programme is below, subject to change.
The language of the event is English.
9:00 - Szabolcs Szolnoki, Ministry of Economic Development: Welcome address
9:10 - Bence Tóth, National Research, Innovation and Development Office: Welcome address
9:20 - András Benczúr, MILAB: Welcome address
9:30 - Dr. Józsa Csaba, NOKIA
9:45 - Attila Bíró, IT ware: Machine vision case studies from geo-distributed collabs and sports
10:00 - Salló György, Qamcom
10:15 - Norbert Barankai, Kincsinfo
10:30 - Gábor Kovács, HUN-REN Alfréd Rényi Institute of Mathematics
10:45 - Coffee Break
11:15 - Bea Benkő, HUN-REN SZTAKI
11:30 - TBC SZTE, TBC
11:45 - András Kalapos, BME, TMIT, SmartLabs: Whitening Consistently Improves Self-Supervised Learning
Self-supervised learning (SSL) has been shown to be a powerful approach for learning visual representations. It allows for learning high-quality representations from unlabelled data, which can be transferred to a wide range of downstream tasks. In this study, we propose to introduce ZCA whitening as the last layer of the encoder in self-supervised learning to improve the quality of the learned features. We show that this simple addition improves performance for a variety of SSL methods across multiple encoder architectures and datasets. Our experiments show that whitening is capable of improving linear and k-NN probing accuracy by 1-5%. Additionally, we propose metrics that allow for a comprehensive analysis of the learned features, provide insights into the quality of the representations and help identify collapse patterns.
12:00 - Domokos Kelen, HUN-REN SZTAKI: Automatic Root Cause Analysis using Asymmetric Shapley Values
Root cause analysis (RCA) is essential for identifying the underlying reasons for problems or failures in various systems. Automated RCA uses advanced algorithms to quickly and accurately determine these root causes, assisting both automated processes and human operators in diagnosing issues. In this presentation, I will discuss our collaboration with Ericsson on automated RCA, employing Asymmetric Shapley Values (ASV). ASV is an extension of the popular SHAP (SHapley Additive exPlanations) framework, designed to handle the complexities of real-world problems where variables have causal relationships that classic SHAP cannot account for. ASV enhances the SHAP framework by incorporating known causal relations, providing more accurate and relevant explanations. Our research has developed a practical methodology for calculating conditional ASV, ensuring that the explanations are theoretically correct under certain conditions. This approach significantly improves the reliability of ASV explanations, making it a valuable tool for identifying root causes in complex systems.
12:15 - Viktor Kovács, Martin Bánóczy, HUN-REN TK, poltextLAB: Nagy nyelvi modellek finomhangolásának optimalizálása
12:30 - Lunch Break
13:30 - László Czúni, Richárd Rádli, Zsolt Vörösházi, University of Pannonia: Veszprém Problems and solutions for fast adaptation in visual object classification
The presentation aims to discuss some major problems in the application of deep neural models for visual classification tasks. Our focus is on such solutions which can be useful in case of few-shot learning with fast adaptation abilities. In our presentation we show two approaches: the usage of fast-learning randomized networks and metric learning. In the first case the deepening of shallow randomized learning models is presented, while for metrics learning a method for the usage of high-level visual or textual cues are proposed to improve precision of the models.
13:50 - Hirling Dominik, Péter Horváth, SZBK: Down the Rabbit Hole: Segmentation Metric Misinterpretations in Bioimage Analysis
In today's scientific environment, with an increasing attention on AI solutions for imaging problems, a plethora of new image segmentation and object detection methods emerge. Thus, quantitative evaluation is crucial for an objective assessment of algorithms. Often, object detection and segmentation tasks utilize evaluation metrics with the same name, but a different meaning due to the differences between object-level and pixel-level classification or just because multiple interpretations coexist. One could argue that in most cases, the meaning should be clear from the context, however, specific and often non-detailed characteristics of the circumstances (e.g. small variations of the task) can make it hard for the readers to understand the exact meaning of different metrics. My presentation is focusing on the various interpretations that have emerged in the research communities related to some segmentation scores. As such, we could identify 5 different definitions for the “average precision (AP)”, and 6 different interpretations for the “mean average precision (mAP)” metrics in the literature. To make things even more complicated, even when some methods work with the same dataset, the metrics used for the evaluation of performance are not necessarily the same. The aims of my presentation are to shed light on some of the main issues with the current state of segmentation and object detection metrics, and to investigate the reasons for the ambiguous use of classification concepts. I’m also going to point out the problems of using similar metrics with nuanced differences by evaluating the 2018 Data Science Bowl (DSB) and 2021 Sartorius neuron segmentation challenge submissions with metrics of similar meaning but slightly differing interpretations.
14:10 - Tamás Szirányi, Marcell Golarits, HUN-REN SZTAKI: 2D és 3D kép/videó/pontfelhő minőségének becslése
14:30 - Levente Hajder, ELTE: Geometric Computer Vision, GCVG@ELTE
The Geometric Computer Vision Group (GCVG) at the Eötvös Loránd University principally deals with 3D machine perception from camera images and LiDAR-scanned 3D point clouds, however, the processing of other sensors; including IMUs, 2D Lidars, ultrasonic sensors, RTK-GPS, microphone arrays; is also addressed. The main focus of the application is autonomous driving. The group has two test vehicles, a car and a controllable go-kart, they can be equipped with the selected sensors. The most important research area of the group is the application of affine transformations for stereo vision. The second focused topic is Lidar-camera and 2D lidar camera calibration for which they use spherical and cylindrical objects instead of the most common chessboard planes. The group strongly cooperates with the Robert Bosch company.
14:50 - András Hajdú, Debrecen
15:10 - Kávészünet
15:40 - András Majdik, Sándor Gazdag, HUN-REN SZTAKI
16:00 - János Gulyás, ELTE: Keep Gesturing: Emerging gesture-based pragmatic communication
”Keep Gesturing” is an innovative game for Extended Reality environments, demonstrating a new dimension of interaction between humans and LLMs through gesture-based communication. Players collaborate with an LLM-controlled avatar to identify and correct discrepancies in an augmented reality environment, relying solely on nonverbal cues.
16:15 - Péter Pollner, ELTE TTK: Synergy between histopathology and proteomics, the case of ovarian cancer
We show that integrating H&E-stained Whole Slide Images (WSIs) with proteomics measurements significantly enhances the prediction of the success of platinum based chemotherapies. We utilize state-of-the-art multi-modal deep learning models for this task. The model is able to better predict overall patient survival as well. Through model interpretability analysis we explore features for highlighting spatial distribution of pathway activities related to platinum treatment response. Our analysis showcases an example where artificial intelligence can assist personalized cancer treatment and suggests some hints for therapeutic vulnerabilities.
16:30 - Dániel Hadházi, BME MIT, MI kutatócsoport: Semiautomatic pulmonary aorta, heart and diaphragm segmentation (Szemi-automatikus pulmonáris aorta, szív és diafragma szegmentálása)
Segmentation of various organs and body parts based on CT scans is an essential step in the computer-aided design of the surgeries. Although, with the emergence of the machine learning-based algorithms many segmentation tasks can now be solved automatically, there is room for semiautomatic algorithms. On the one hand, in certain circumstances -- such as poor quality of the scans, presence of artifacts, and in some corner cases -- even machine learning based methods fail to perform the segmentation accurately. In these cases, a robust semiautomatic algorithm may be used as a fallback mechanism. On the other hand, training datasets which typically cover a wide spectrum of cases and range of qualities for a better generalization ability of the trained model, require however time consuming manual labeling. Furthermore, a slightly different application requirement may require the re-labeling of the data. Semiautomatic algorithms can be a viable solution for speeding up the manual labeling of the training datasets. Finally, there are practical tasks that have not yet been solved by automatic segmentation methods -- such as diaphragm detection -- that healthcare professionals can benefit from.
16:45 - Csaba Kerepesi, HUN-REN SZTAKI: Measuring biological age by artificial intelligence
Aging clocks are artificial intelligence models that can estimate the age of an individual. However, biological age can differ from chronological age: a value higher or lower than chronological age may indicate accelerated or slowed aging. We develop and apply aging clocks for better understanding of biological aging and measuring biological age based on different types of data. The developed tools may become suitable for the evaluation of potential rejuvenating therapies or interventions that can slow down the aging process. In addition, they could become common tools for personalized medicine also considering the biological age of the patient.
17:00 - Closing Remarks