Auto AI Seminar (WP7)
The WP7 seminar series consists of two parts: a scientific and a social part. Below you can find a detailed overview of each part for the first seminar on the 9 November from 15:30-17:00.
________________________________________
Scientific part: Zoom meeting
15:30-16:00 Blaž Škrlj : “Automating text classification via representation evolution”
Abstract
Learning from less structured data sources, such as texts, enables the construction of systems capable of sentiment analysis, recommendations, bot detection and more.
In recent years, large neural language models have shown great promise for this task, albeit being computationally more intensive and offering fewer insights into what was actually learned.
We first present the problem of text classification, its properties and other aspects relevant when designing an autoML system. Next, we present a possible solution to some of the mentioned problems, which incorporates neuro-symbolic representation learning. We demonstrate that considering both symbolic and sub-symbolic representations of documents, better-than-symbolic models can be obtained whilst maintaining most of the interpretability. Further, we explored the transferability of solutions across different text classification tasks, demonstrating that representation-level information is potentially very useful as a prior when addressing a new task.
16:00-16:20 — Round table updates by all WP7 participants
16:20-16:30 — Find a (recurring) slot for upcoming TAILOR WP7 seminars
16:30-17:00 Social part: GatherTown
• Suggestions for improving the seminar format
• Informal chats about possible collaborations
• Brainstorm/discuss mechanism to track WP7 publications info (including collaboration between partners etc)
• Chat about anything else you feel worthwhile!
________________________________________
Hope to see you at the seminar!
Best regards,
Adem, Holger, Koen and Manuel (the TAILOR WP7 Seminar Organising Committee)