WORKSHOP: Model Collapse

! ENG translation below !

𝗙𝗲𝗹𝗶𝗰𝗶𝘁𝘆 𝗛𝗮𝗺𝗺𝗼𝗻𝗱
𝐾𝑜𝑙𝑎𝑝𝑠 𝑚𝑜𝑑𝑒𝑙𝑜𝑣
DELAVNICA | 1. april 2026, 16:00-19:00 | ALUO, Tobačna 3
✦ aksioma.org/becomingimage/sl/delavnice/kolaps-modelov/

✦ 𝗨𝗱𝗲𝗹𝗲𝘇̌𝗯𝗮 𝗷𝗲 𝗯𝗿𝗲𝘇𝗽𝗹𝗮𝗰̌𝗻𝗮, 𝗼𝗯𝘃𝗲𝘇𝗻𝗮 𝗷𝗲 𝗽𝗿𝗶𝗷𝗮𝘃𝗮! → pretix.eu/aksioma/becomingimage/ ✦

V tej praktični delavnici bodo udeleženci eksperimentirali z ustvarjanjem razširjenih kolažev kot orodjem za razumevanje kroženja podob v dobi strojnega učenja. Raziskovali bodo pojav kolapsa modelov: degenerativni učni proces, ki nastane, ko se model uri na vsebinah, ustvarjenih z umetno inteligenco. Podobe umetne inteligence obračajo indeksikalno logiko fotografije in imajo vlogo pri oblikovanju sveta; skozi delavnico se bomo spraševali, kaj se zgodi, ko podoba predhodi resničnost.
Čeprav raziskujemo orodja strojnega učenja, v tej delavnici ne bomo uporabljali generativnih orodij umetne inteligence.

Cilji delavnice: udeleženci bodo razvili razumevanje, kako:
– kolaps modelov povzroča izgubo podatkov, degradacijo in abstrakcijo znotraj generativnih sistemov umetne inteligence,
– imata kolaž in generativna umetna inteligenca številne prekrivajoče se lastnosti,
– je označevanje podatkov omejujoča in neetična praksa.

🟢 SORODNI DOGODEK:
RAZSTAVA: 𝐿𝑎𝑡𝑒𝑛𝑡𝑛𝑖 𝑝𝑟𝑜𝑠𝑡𝑜𝑟
Otvoritev: 1. april ob 20:00 | Aksioma projektni prostor

—

V okviru 𝒕𝒂𝒌𝒕𝒊𝒌𝒆&𝒑𝒓𝒂𝒌𝒔𝒂 #17: 𝑷𝒐𝒔𝒕𝒂𝒋𝒂𝒏𝒋𝒆 𝒑𝒐𝒅𝒐𝒃𝒂, ki ga kurira Marco De Mutiis.
aksioma.org/becomingimage/sl

[ENG]

𝗙𝗲𝗹𝗶𝗰𝗶𝘁𝘆 𝗛𝗮𝗺𝗺𝗼𝗻𝗱
𝑀𝑜𝑑𝑒𝑙 𝐶𝑜𝑙𝑙𝑎𝑝𝑠𝑒
WORKSHOP | 1April 2026, 16:00-19:00 | ALUO, Tobačna 3
✦ aksioma.org/becomingimage/workshops/model-collapse/

✦ 𝗙𝗿𝗲𝗲 𝗮𝗱𝗺𝗶𝘀𝘀𝗶𝗼𝗻, 𝗿𝗲𝗴𝗶𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝗱! → pretix.eu/aksioma/becomingimage/ ✦

In this practical workshop, participants will experiment with expanded collage making as a way to understand the circulation of images in the age of machine learning. They will explore the phenomenon of model collapse: a degenerative learning process caused when a model is trained on AI generated content. AI images reverse the indexical logic of photography and play a role in shaping the world; through this practical workshop, we will ask what happens when the image comes first.
Whilst we are exploring machine learning tools, we will not be using generative AI software in this workshop.

Takeaways: participants will develop an understanding of how:
– Model collapse causes data loss, degradation, and abstraction within generative AI systems.
– Collage and generative AI have many overlapping qualities.
– Data labelling is a limiting and unethical practice.

🟢 RELATED EVENT:
EXHIBITION: 𝐿𝑎𝑡𝑒𝑛𝑡 𝑆𝑝𝑎𝑐𝑒
Opening: 1 April at 20:00 | Aksioma Project Space

—

In the framework of 𝒕𝒂𝒄𝒕𝒊𝒄𝒔&𝒑𝒓𝒂𝒄𝒕𝒊𝒄𝒆 #17: 𝑩𝒆𝒄𝒐𝒎𝒊𝒏𝒈 𝑰𝒎𝒂𝒈𝒆 curated by Marco De Mutiis.
aksioma.org/becomingimage/