Schlagwort: NLP

Our next Expert Meeting on Monday, 21 August 2023, 17:00-18:30, will focus on various aspects and applications of Large Language Models.

It will take place at the ZHAW premises in Lagerstrasse 45, 8004 Zurich in room ZL O3.01 on the third floor. The meeting will be followed by an apéro. Online participation is also possible.

Please use the form to confirm your attendance by July 21st: https://forms.gle/44BUEBjKrVuW4WLV6 

We will then send you a calendar invitation which includes online participation details.

The following presentations are confirmed for the meeting:

Kim Engels, Converto AG – Large Language Models for Cross-Media Marketing

In this talk, Kim briefly presents some examples of how he and his team at Converto use AI and LLM to improve or speed up their projects.

Besides approaches such as text generation for newsletters, there are also variants such as code generation within the team as well as the use of self-developed solutions to create 3D models for customer campaigns. 

Florian Tramér, ETH – Are Aligned Neural Networks Adversarially Aligned?

Large language models are now tuned to align with the goals of their creators, namely to be „helpful and harmless.“ These models should respond helpfully to user questions, but refuse to answer requests that could cause harm. However, adversarial users can construct inputs which circumvent attempts at alignment. In this talk, we’ll discuss to what extent these models remain aligned, even when interacting with an adversarial user who constructs worst-case inputs (adversarial examples). We’ll see that existing optimization attacks are insufficiently powerful to reliably attack aligned text models, except when these models are multimodal (i.e., they can process both text and images). In that case, we show these models can be easily attacked, i.e., induced to perform arbitrary un-aligned behavior through adversarial perturbation of the input image.

Alex Paramythis, Contexity AG – Adapting Large Language Models for Customer Request Handling

With the rise of Generative Large Language Models (LLM), companies are looking into the many opportunities proffered by this new technology. One area of particular interest is the automated handling of customer requests (e.g., received through email, chat, social media, etc.) using the institutional knowledge at hand. In such a context, LLMs may need to be trained on, or have access to, privileged, non-public information in the company’s knowledge base. This, in turn, entails that the models need to be prepared within, and served from, a company’s own infrastructure to prevent information leakage — a requirement that points in the direction of commercially friendly open-source models. In this talk we will present our work on generation of responses to customer requests using the IGEL (a BLOOM based model), FLAN-UL2, and Falcon LLMs. For the first two models we will also report on our attempts to fine-tune the models before use, with a variety of training data.

This Expert Meeting will take place at the ZHAW premises in Lagerstrasse 45, 8004 Zurich in room ZL O3.01 on the third floor (online participation is also possible for those who prefer this option) on Wednesday, May 10 from 17:30-19:00. After the meeting, there will be an apéro so that you can carry on your discussions and get to know each other.

We have the following two talks confirmed:

End-to-end ASR for Swiss German at Microsoft: A Transducer Approach
Oscar Koller, Applied Scientist at Microsoft

Automatic speech recognition (ASR) for Swiss German is a challenging task due to the lack of a standardized writing system and the high regional variability of the dialects. In this talk, we present our work on developing end-to-end ASR models for Swiss German at Microsoft using transducer architectures. We show that transducers outperform hybrid models by over 20% in word error rate on a multi-dialectal corpus of Swiss German speech. We also compare our models to Whisper, a state-of-the-art sequence-to-sequence model for low-resource ASR, and find that transducer models achieve comparable results with much smaller model size and training time. Finally, we discuss how end-to-end models produce transliterations of Swiss German words instead of standard German translations affecting the readability and usability of the output and propose solutions to this problem.

Revolutionizing Natural Interaction with Swiss German: A Glimpse into the Future of Conversational AI
Claudio Paonessa and Yanick Schraner, Researchers at FHNW

Get ready for a glimpse into the future of natural interaction with computer systems in Swiss German! We leveraged the latest advancements in speech-to-text and text-to-speech technology to create an engaging and interactive experience that showcases the results of our cutting-edge research.

Exploring the Acceptance of Intelligent Voice Assistants in Home Care Applications: Opportunities and Obstacles [10 mins presentation, 10 mins discussion]
Edith Birrer, Researcher at iHomeLab – HSLU (Hochschule Luzern)

In the scope of co-creation sessions, care workers provided insights on applications and on concerns about Intelligent Voice Assistants (IVA) in the home of their clients or patients. The sessions focused on the potential to support the care documentation process by IVA. Participants’ expectations and worries spanned from the ability to handle dialects, to confidentiality issues, to integration in existing care documentation systems. However, there is a general openness toward the idea to employ IVA as means to improve the quality of care. The challenge foreseen for using IVA is to become as time efficient as care documentation systems in place. Alternatively, as suggested by participants, IVA could complement existing processes or even create new ones in the care context.

If you want to join, please fill in the following registration form by April 27: https://forms.gle/PmRQENtY8aybJeby5
Please note that the registration form includes information for the SwissNLP General Assembly which is co-located.

It is our pleasure to invite you to an upcoming event organised by the data innovation alliance and SwissNLP, the Swiss Association for Natural Language Processing (https://swissnlp.org/) on Tuesday, May 3rd at Nüü (Lagerstrasse 45, 8004 Zurich, ground floor).

We will combine an NLP Expert Group Meeting on Multimodal AI with the SwissNLP General Assembly 2022, followed by an apéro and socialising/networking.

• 16:00 – 17:00: SwissNLP General Assembly 2022
  • Annual and Financial Report 2021
  • Budget & activities for 2022
  • Board elections for 2022
  • Discussions & Varia
    
• 17:15 – 18:30: Expert Group Meeting on Multimodal AI
  An exciting new research direction, Multimodal Artificial Intelligence concerns the creation of AI models which can jointly process different types of inputs such as images, text, audio, video, or structured (tabular or graph) data. This opens up possibilities for applications which can interact with the world in new ways and address much more complex use cases.

A recent example is MUM [1], Google’s “Multitask Unified Model”, which can answer multimodal search queries such as “Are these hiking boots [see picture] suitable to hike Mount Fuji?”

In the meeting, we plan for 1-2 short presentations (10min each), followed by an open discussion of the topic. One potential (and intended) outcome is ideas for joint projects!

• Looking for Speakers: If you would like to share your experience in Multimodal AI with the group, please let us know. The presentation may include a success story, a collection of best practices, an open problem, or even an outline of a project that would benefit from Multimodal AI. It would be great to discuss positive experiences, as well as blockers and limitations. If you would like to present, please send us your presentation titles (email to ciel@zhaw.ch, natalia.korchagina@pwc.ch). Deadline is Monday, April 25.
  
• From 18:30: Apéro & socialising/networking
  Join us to exchange & mingle over drinks & snacks!

It would be great to see many of you in person, but online participation is also possible for both events.

Please register using the link from the invitation mail.
Deadline for registration: Wednesday, April 27.

Topic: Swiss German Speech Processing
Date: Tuesday, November 9, 2021
Time: 17:00
Location: online or on premise in Zurich (will be decided based on preference of participants)
Registration: until November 1^st

Please register until the 5th of January for the meeting.

If you’re interested in sharing your experiences as a speaker contact either of the leaders until December 18th.

More details about the meeting can be found in the member area!