Tech-ai

Proud to share a significant research achievement in the field of Artificial Intelligence!

The project "AI for All: Unlocking Scalable and Democratized Foundation Models," led by Prof. Assaf Schuster from the Technion, has been selected as part of a new VATAT (Planning and Budgeting Committee) program. This initiative is dedicated to encouraging the establishment of groundbreaking "Moonshot" AI research hubs in Israeli research universities.
The program aims to promote ambitious research with the potential for broad impact, and to further strengthen Israel's position at the forefront of global AI research.

Congratulations to Prof. Schuster and the research team on this remarkable achievement!

🚨 ההרשמה נסגרת – וזו ההזדמנות האחרונה! 🚨

רובוטים כבר לא מדע בדיוני – הם כאן, והם משנים את העולם.
ב־7-8 בספטמבר בטכניון מתקיים קורס מערכות אוטונומיות של רובוטים עם פרופ' קיריל סולוביי, מהמובילים בעולם בתחום.

👩‍💻 מתאים למהנדסים, אנשי פיתוח וגם לכל מי שסקרן – אין צורך בניסיון קודם.

⏰ אל תפספסו – ההרשמה נסגרת בקרוב!

👉 לפרטים נוספים והרשמה: https://lp.vp4.me/16ab

הצטרפו לאירוע הכי מרתק בקיץ הקרוב!

Home | iSpeech-25 iSpeech25 Conference Technion - Israel Institute of Technology AI Hub conducts the second Israeli conference on Speech & Audio processing using neural nets. It is a venue for presenting the most recent work on the science and technology of spoken language processing both in academia and industry

כנס iSpeech - ההרשמה נפתחה!
הצטרפו אלינו ב-23.7.25 לשמוע את פריצו הדרך בתחום הבינה המלאכותית, מולי השפה וטכנולוגיות הדיבור!
https://www.techaiconf.org/

עולם ה-Graph Search מרתק ורלוונטי מתמיד, ומשפיע רבות על תחומים כמו רובוטיקה, בינה מלאכותית, וניתוח רשתות.
בקורס חדש אשר יתקיים ביולי הקרוב בטכניון, תלמדו כלים פרקטיים לעולמות תוכן רבים, ותוכלו ליישמם בקלות בעבודה.
פרטים נוספים והרשמה כאן: https://lp.vp4.me/19q8

שמרו את התאריך:
כנס iSpeech יתקיים ב-23.7.25 בטכניון
פרטים נוספים והרשמה - בקרוב!

** AI for Smarter Cities **
The way we design and analyze urban spaces is evolving, and AI is at the forefront of this transformation. By leveraging Machine Learning (ML) tools and Earth Observation (EO) data, we can gain deeper insights into city structures, enhance urban planning, and develop adaptive strategies for the future.
In this field, Coral Hamo Goren is making a significant impact. Coral is a PhD student at the Technion’s Faculty of Architecture and Town Planning and conducting groundbreaking research under the guidance of Prof. Guy Austern and Prof. Yasha (Jacob) Grobman.
In her recent ACADIA 2024 publication, she introduced a Faster R-CNN_ResNet50 model trained to detect and classify building typologies. Expanding on this, her follow-up research moves from individual buildings to the urban block level, integrating heuristic algorithms, supervised learning, and K-Means clustering to model spatial organization. This approach ensures its applicability across diverse urban environments and supports data-driven city planning.
Beyond her academic achievements, Coral was awarded the prestigious Gutwirth Scholarship, is a proud mother of three, and served in the reserves during the Swords of Iron War as an operations officer in Givati! 👏
Read her full ACADIA 2024 paper here: https://shorturl.at/j8cfQ

Spots are limited! Register today!

Robot Autonomy - קורס פרופ' קיריל סולוביי

🔬AI in Medical Imaging: Transforming Cancer Diagnostics & Treatment

🧪Translational science has propelled significant advancements in AI within medicine, particularly in the development of real-world healthcare applications to improve patient outcomes.

Prof. Yonatan (Yoni) Savir, healthcare researcher at the Technion - The Ruth and Bruce Rappaport Faculty of Medicine, and the Technion's translational AI hub, Tech-ai, and member of the Zimin Institute for AI Solutions in Healthcare, shared with HIT Consultant Media how AI-aided data integration from various sources such as medical images, molecular data, and medical records can provide better cancer diagnostics and treatment planning.

Zimin Foundation

Exciting News from the Technion-Rambam Initiative in Medical AI (TERA)!
We are thrilled to announce that Anat Rotschield has been awarded a prestigious TERA PhD Scholarship! Anat , a PhD student at Technion, has been passionate about research from a young age, fascinated by how small biological mechanisms can have a big impact on human health. After earning both her bachelor's and master's degrees at the Technion, Anat developed a strong interest in bioenergetics and cardiac research.
Now, as a TERA PhD Scholar, Anat is focusing on the bioenergetics of atrial myocytes and the triggers of atrial fibrillation after bypass surgery—harnessing the power of AI to advance cardiac care. 💙 Outside the lab, she enjoys cooking, reading, running, and traveling.
Huge congratulations to Anat who will research under a supervision of Prof. Yael Yaniv - Professor at Biomedical Engineering Faculty, and co-supervision of Dr. Tom Friedman, Head of the Cardiac Surgery research Unit, Department of Cardiac Surgery at Rambam Health Care Campus. 👏

Graph Neural Networks (GNNs) are the leading method for learning on graph-structured data, primarily relying on the Message Passing (MP) paradigm, in which messages are passed locally between nodes in order to learn meaningful node or graph representations. However, MP-based models have limited expressive power, for example, they will always have the same output on any 2-regular graph. Subgraph GNNs, a new class of graph architectures in which MP is performed on several subgraphs of the original graph, improve expressiveness (and accuracy) but often come with high computational costs, making it challenging to operate on larger graphs.

In their latest NeurIPS paper, Guy Bar-Shalom, Yam Eitan (co-first author), Dr. Fabrizio Frasca, and Prof. Haggai Maron present “A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening”, which enables a fine-grained control over the trade-off between expressivity and efficiency. This allows for the processing of larger graphs while still retaining the expressivity benefits of Subgraph GNNs.
Their work extends a key insight from their previous research: Subgraph GNNs can be reformulated as message passing on the graph Cartesian product. They relax one element in the product by first applying a graph clustering algorithm to coarsen one copy before computing the product, allowing precise control of computational complexity. Notably, this leads to a new object—a product of a graph and its coarsened version—that reveals unexplored symmetries. Using geometric deep learning principles, they characterize these invariances and design neural network layers that respect them.

Key Findings:
* Significantly better accuracy compared to other efficient Subgraph GNNs.
* Scales to larger graphs while maintaining expressivity benefits.
* Matches the performance of original Subgraph GNNs when maximizing expressivity (at the cost of efficiency).
* Studying a new and interesting symmetry structure.

Notably, this work was also awarded **Best Paper** at NeurReps – Symmetry and Geometry in Neural Representations, a NeurIPS workshop! 🎉

Curious to dive deeper? Read the full paper: https://openreview.net/pdf?id=9cFyqhjEHC

💫 Another Spotlight on AI Research at the Technion! 💫
Generalization in deep neural networks remains a fundamental mystery—despite their ability to fit even random labels, they often generalize well in practice, seemingly in defiance of the classical theory of generalization.
This raises the question—how well do neural networks generalize when trained to perfectly fit a noisy dataset?
In a recent paper published at NeurIPS 2024, Itamar Harel, William M. Hoza (University of Chicago), Gal Vardi (Weizmann Institute of Science), Itay Evron, Prof. Nati Srebro (Toyota Technological Institute at Chicago), and Prof. Daniel Soudry consider fully connected neural networks with binary weights, analyzing both minimal networks (with the fewest possible weights) and typical networks (randomly initialized ones that interpolate the data), proving that in both cases generalization error is proportional to the noise in the data, similar to empirical evidence. Unlike previous work, their results apply to overparameterized networks with input dimension that is neither very large nor very small.

You can read the full article here: https://arxiv.org/abs/2410.19092