Sunday 10 September: Registration & Accommodation of Participants
Monday 11 September: Opening and the Impact of GAIM
Online session: Google Meet acces link
9:00: Opening session (moderated by Dan Cristea):
- Natalia Gașițoi - Rector of the "Alecu Russo" Bălți State University
- Valentina Prițcan - Vice Rector of the "Alecu Russo" Bălți State University
- Dan Tufiș - Director of the AI Research Center, Romanian Academy
- Vasile Apopei - Director of the Institut of Computer Science, Iași branch, Romanian Academy
- Adrian Iftene - Dean of the Faculty of Computer Science, "Alexandru Ioan Cuza" University of Iași
- Mircea Petic - Organising Committee
9:45: Coffee break
10:30: Mădălina Cocea (chaired by Daniela Gîfu)
Practical session: AI Tools as Research Assistants. A Practical Guide to Use ChatGPT for Writing Scientific Papers. (Presentation)
A presentation on AI-based tools useful to the academic environment; the strategic use of ChatGPT for the verification of information; the regulations of major scientific editing houses regarding the use of these tools.
13:00: Lunch
14:30: George Tecuci (chaired by Dan Tufiș)
Tutorial: Amazing Applications of CogentGPT (Presentation)
This tutorial explores the amazing potential applications of CogentGPT, an envisioned system that integrates the learning and reasoning capabilities of Cogent (an instructable agent for cogent analysis), with the natural language processing capabilities of ChatGPT, (a Chat Generative Pre-training Transformer).
These applications include: automatic fake news detection (such as, the Ghost of Kyiv), automatic critique of analysts’ draft reports (by highlighting additional relevant evidence, and identifying strengths and weaknesses in reasoning), automatic evaluation of students’ arguments, and automatic query obfuscation.
16:00: Tea break
16:30 - 18:00 City tour
19:00 Welcome party & special guest session: Nancy Ide (chaired by Dan Cristea)
Personal Remarks on the Evolution of the Domain
Tuesday 12 September: Applications of GAIM
Online session: Microsoft Teams access link
9:00 - 10:30 Tudor Toma (chaired by Dan Cristea)
Tutorial: AI for Enhanced Respiratory Medicine: Practical Applications and Potential in Healthcare (Presentation)
This lecture will explore the gap (and the bridge we can build) across clinical practice and cutting-edge AI technology, particularly in the realm of respiratory medicine. It will discuss potential ways to apply AI systems to enhance medical practice, with a focus on respiratory medicine as a case study. Topics covered will include enhancing doctor to doctor communications as well as doctor to patient through NLG, assisting in diagnosis and treatment and automatisation of interventional procedures, plus collaboration and future directions. The lecture will be followed by extensive interactive discussions and a Q&A session that will foster brainstorming new ideas and research possibilities to create impactful solutions applicable across various medical areas.
10:30 Coffee break
11:00 - 12:30 Vasile Păiș (chaired by Diana Trandabăț)
Tutorial: Legal Domain Named Entity Recognition (Presentation)
In natural language processing (NLP), named entity recognition (NER) is an important process towards information extraction and text understanding. Domain-specific named entities (NEs) are particularly important and harder to detect compared to more general ones (such as the traditional person, organization, location).
As part of this talk, I will cover recent developments regarding legal-domain NER, as well as recent activities focusing on the Romanian language, such as the LegalNERo and MicroBloggingNERo corpora.
We will explore the annotation process and automatic systems using the RELATE platform, dedicated to the Romanian language. Generative AI systems can be used for data augmentation purposes, in order to increase the performance of the NER systems.
I will cover the NL-Augmenter framework, which includes Generative AI for NER and other tasks. Other Generative approaches for corpus building and augmentation with direct impact on NER will be described as well.
13:00 Lunch
14:30 - 16:00 Tudor Toma and Valentin Nagacevschi
Practical session: Prompt Engineering - a Practical Guide
I will present use cases to show the meaning of prompt engineering the anatomy of a prompt, how prompting can improve the response of a GAIM system
16:00 Tee break
16:30 - 18:00 Vasile Păiș
Practical session: Using RELATE and NL-Augmenter
The students will interact with the RELATE platform's corpus building component and will use the NL-Augmenter's generative capabilities to generate new sentences for data augmentation.
19:00 Dinner & social activities
Wednesday 13 September: How Could GAIM be Useful
Online session: Microsoft Teams access link
9:00 - 11:00 Florin Olariu (chaired by Adrian Iftene)
Tutorial: Introducing GitHub Copilot: Your Brilliant AI Pair Programmer! (Presentation)
Coding becomes an exceptional experience when you have a Copilot by your side. This ground-breaking AI tool, driven by the incredible GPT-3.5, turns your coding adventure into a smooth, collaborative adventure.
Embrace the future of programming as Copilot learns from your projects, understands your coding style, and predicts your coding needs like a genuine coding partner.
Gone are the days of sifting through code snippets or battling with sophisticated algorithms. Copilot creates ideas, completes lines of code, and even provides full code blocks based on your intent.
Say farewell to time-consuming code exploration and embrace the swift, intuitive coding flow provided by GitHub Copilot. Supporting multiple programming languages, this intelligent assistant adapts to your preferred environment, saving you valuable time and effort.
Unlock your coding potential with GitHub Copilot — a game-changer that empowers you to bring your creative ideas to life, revolutionizing how you write code.
Elevate your programming endeavors and embark on a journey where innovation meets artificial intelligence, all within the realm of GitHub Copilot, your extraordinary AI pair programmer!
Practical session: Embrace Thrilling Challenges and Leverage Inspiring Code Samples by Using GitHub Copilot!
GitHub Copilot emerges as your ideal ally in the world of .NET code, altering the way you approach projects.
Prepare to go on a coding adventure like no other, as Copilot enables you to handle hard tasks with ease.
11:00 Coffee break
11:30 - 13:00 Cristian Simionescu
Webinar: BrainFuse: Generative AI applied to Medical Imaging (Presentation)
We will discuss the motivation, risks and impactful considerations in trying to apply GAIM’s to medical Imaging. We will do so by trying to build such a method.
13:00 Lunch
14:30 - 16:00: Paola Monachesi (chaired by Dan Cristea)
Tutorial: Impact of AI on Social Media: ChatGPT and Stable Diffusion (Presentation)
AI generative models will have an impact on the future of social media and creative industries since they can contribute, among other things, to content creation. We will consider the role of two relevant models in this context, such as Stable Diffusion and ChatGPT.
The former is a model able to generate photo realistic images given any text input while the latter is a language model able to generate text, based on context and past conversations. These tools can support creativity and innovation in developing new content, but can also be a source of biases due to the uncurated training data on which they are based. Examples will be provided on how Stable Diffusion further amplifies biases with regards to the portrayal of women.The images generated by the Stable Diffusion model are considerably more explicit than those in the underlying training data for the same semantic concepts. It is unclear why Stable Diffusion amplifies these biases.
AI-enhanced social media poses a clear threat since the flow of high-quality deep fake into social media could increase and invade public conversation by making citizens disoriented, distrustful and angry.
16:00 Tee break
16:30 - 18:00 Paola Monachesi
Practical session: Hands-on.
In the practical session, students can choose among several projects depending on interests and expertise, such as: 1. develop social media content to promote a Sustainable Development Goal by using Stable Diffusion and ChatGPT or 2. detect biases in the webversion of Stable Diffusion and propose solutions or 3. compare training data and Stable Diffusion results in the production of biases and hypothesize solutions.
19.00 Dinner & Social activities
Thursday 14 September: In the Very Heart of GAIM
Online session: Microsoft Teams access link
9:00 - 10:30 Dragoș Gavriluț (chaired by Adrian Iftene)
Tutorial: LL-LLM (Low Level Large Language Models) - a Deep Dive into how Large Language Models Work (Presentation)
During this lecture we will discuss how large language models are being build (starting from the how text is being converted from a sequence of characters into the input of a neural network, and continuing to how transformers and encoders/decoders are being build).
10:30 Coffee break
11:00 - 12:30 Andrei-Marius Avram (chaired by Dan Tufiș)
Tutorial: Speech Recognition in the Large Language Models Era (Presentation)
The advent of large language models, such as Wav2Vec2 and HuBERT, has revolutionized the field of speech recognition, unlocking remarkable advantages and enhancing the performance of speech-to-text applications. These models leverage self-supervised learning techniques to extract meaningful representations from raw audio, enabling highly accurate transcription even in the absence of large labeled datasets. Moreover, by incorporating generative AI, these models have further improved the quality and fluency of transcriptions, refining the output through context-aware language generation. This tutorial explores the remarkable advantages brought by this technology, shedding light on the transformative impact of large language models and the potential for even greater achievements in the field of speech recognition through the integration of generative AI.
13:00 Lunch
14:30 - 16:00 Dragoș Gavriluț
Practical session: Excessing BPE
In this practical session we will try building a BPE algorithm (both training and various optimisations) with different kinds of encodings (Unicode / Ascii or UTF-8). We will explore how tokens are being built, as well as different ways to pre-process a text before tokenizing it.
16:00 Tee break
16:30 - 18:00 Andrei-Marius Avram
Practical session: Advancing Romanian Speech Recognition: Fine-tuning LLMs and Harnessing Generative AI
In this practical session, we explore the fine-tuning of an audio LLM for Romanian Speech Recognition. Additionally, we investigate the utilization of generative AI to improve transcription results, showcasing the potential synergy of these technologies in the field of language processing.
19.00 Dinner & Social activities
Friday 15 September: How to Make GAIM More Logical, Webinar and Closing
Online session: Microsoft Teams access link
9:00 - 10:30 Ștefan Trăușan-Matu (chaired by Dan Cristea)
Tutorial: The integration of generative AI and Knowledge Graphs. Why needed, existing resources, and possible approaches (Presentation)
Generative AI achieved amazing performances in recent years, the best example being ChatGPT which, like similar systems, is based on Large Language Models built from huge corpora, using Deep Neural Networks (DNN). Even if, generally, it generates coherent text, some limitations were identified, for example in handling logical reasoning and that sometimes it even hallucinates. Other problems faced by DNN and other machine learning techniques are the difficulty of providing explanations (the subject of XAI - eXplanatory AI) and ethics. Therefore, for handling such problems, in addition to corpus resources, containing implicit knowledge, explicit knowledge structures (knowledge graphs, ontologies, and rules) and associated processing approaches are needed.
10:30 Coffee break
11:00 - 12:30 Ștefan Trăușan-Matu and Razvan Paroiu
Practical session: Identifying limitations of ChatGPT and building Knowledge Graphs
The session will have two parts: (1) ontological limitations will be explored during interacting with ChatGPT; (2) an ontology will be built.
13:00 Lunch
14:30 Futurity Workshop (chaired by Daniela Gîfu and Diana Trandabăț)
18:00 - 19:00 Dinner