Dnevi Jožefa Stefana, DANES ob 13:00, področje INFORMACIJSKIH ZNANOSTI, prof. dr. Nicu Sebe

[Martin Klanjsek]

Vabimo vas na 4. znanstveno predavanje, in sicer s področja informacijskih znanosti, v okviru letošnjih Dnevov Jožefa Stefana, ki bo danes, 26. marca 2026, ob 13:00 v Veliki predavalnici Instituta »Jožef Stefan« na Jamovi cesti 39 v Ljubljani. Predavanje bo v angleščini. Na voljo bodo kava, pijača in prigrizki.

 

 

Kako se umetna inteligenca uči videti, ustvarjati in presojati pravično

 

prof. dr. Nicu Sebe

Univerza v Trentu, Trento, Italija

 

 

V prvem delu predavanja bom predstavil, kako učimo umetno inteligenco ustvarjati posnetke brez podrobnih označb ali oznak, vezanih na posamezne objekte. Z učenjem na zbirkah podobnih posnetkov – na primer obrazov ali človeških teles – se model nauči posploševati na celotno skupino primerov. Na tej ideji temelji tudi naš učljivi pogon za igre (LGE iz angleškega learnable game engine), ki se iz preprostih posnetkov z eno kamero nauči slediti prizorom in objektom ter jih ponovno izrisovati iz različnih zornih kotov. Podobno kot pravi pogon za simulacijo iger zajame osnovno fiziko in logiko, kar uporabnikom omogoča poseg v dogajanje ali usmerjanje virtualnih agentov z navodili v naravnem jeziku. Drugi del predavanja je posvečen varnosti in pravičnosti generativne umetne inteligence. Večina obstoječih pristopov zaznava le vnaprej določene vrste pristranskosti, vendar se v resničnih sistemih pojavljajo tudi nepričakovane. Zato predstavljamo OpenBias, metodo, ki odkriva in meri prej neznane pristranskosti v modelih za pretvorbo besedila v sliko, ne da bi se opirala na vnaprej določen seznam vrst pristranskosti. Naši eksperimenti kažejo, da se OpenBias dobro ujema z uveljavljenimi metodami in s človeškimi ocenami, kar omogoča bolj prilagodljivo vrednotenje pravičnosti generativnih sistemov.

 

O predavatelju: Nicu Sebe je redni profesor na Univerzi v Trentu, kjer vodi raziskave na področju iskanja multimedijskih informacij ter interakcije med človekom in računalnikom v aplikacijah računalniškega vida. Je sourednik revije Computer Vision and Image Understanding ter nosilec številnih vodilnih organizacijskih vlog na konferencah, med drugim ACM Multimedia, ICCV, ECCV in CVPR.

 

 

Napovednik predavanja in posnetke preteklih predavanj najdete na spletni strani »Kolokvijev na IJS«  <https://kolokviji.ijs.si/> https://kolokviji.ijs.si/, posnetke pa tudi posebej na  <https://video.arnes.si/?q=Kolokviji_IJS> https://video.arnes.si/?q=Kolokviji_IJS (zadnje štiri sezone) in  <http://videolectures.net/kolokviji_ijs> http://videolectures.net/kolokviji_ijs (starejši).

 

______________

 

 

We invite you to the 4th scientific lecture, dedicated to the field of information sciences, as part of this year's Jožef Stefan Days. The lecture will take place today, 26 March 2026, at 1:00 PM in the main lecture hall of the Jožef Stefan Institute, Jamova cesta 39 in Ljubljana. The lecture will be held in English. Coffee, drinks, and snacks will be provided.

 

 

How AI learns to see, generate, and judge fairly 

 

Prof. Dr. Nicu Sebe

University of Trento, Trento, Italy

 

 

In the first part of the lecture, I will explore how we teach AI to generate videos without relying on detailed annotations or object‑specific labels. By training on collections of similar videos – such as faces or human bodies – the model learns to generalize across an entire category. Building on this idea, we developed a learnable game engine (LGE) that learns from simple monocular videos to keep track of scenes and objects and to re‑render them from different viewpoints. Much like a real game engine, it captures basic physics and logic, allowing users to control the scene or guide virtual agents through high‑level language instructions. The second part of the lecture turns to the safety and fairness of generative AI. Most existing approaches look only for predefined types of bias, but real‑world systems can exhibit unexpected ones. To address this, we introduce OpenBias, a method that uncovers and measures previously unknown biases in text‑to‑image models without relying on any preset list. Our experiments show that OpenBias aligns well with established methods and with human judgment, offering a more flexible way to assess fairness in generative systems.

 

About the lecturer: Nicu Sebe is a full professor at the University of Trento, where he leads research in multimedia information retrieval and human-computer interaction in computer vision applications. He is a co-editor in chief of the Computer Vision and Image Understanding journal and has held numerous leading organizational roles at major conferences, including ACM Multimedia, ICCV, ECCV, and CVPR.

 

 

The lecture trailer and recordings of past lectures can be found on the website of »JSI colloquia«  <https://kolokviji.ijs.si/> https://kolokviji.ijs.si/, and recordings specifically at  <https://video.arnes.si/en/?q=Kolokviji_IJS> https://video.arnes.si/en/?q=Kolokviji_IJS (last four seasons) and  <http://videolectures.net/kolokviji_ijs> http://videolectures.net/kolokviji_ijs (older).

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.ijs.si/pipermail/kolokvij-ijs/attachments/20260326/1cc0ac01/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 112138 bytes
Desc: not available
URL: <https://mailman.ijs.si/pipermail/kolokvij-ijs/attachments/20260326/1cc0ac01/attachment-0001.png>