Software and Resources
Resources
ReProv-API
Description: API-based solution enabling users to register and execute workflows seamlessly by integrating with the REANA execution system.
Abstract: ReProv-API is an API-based solution enabling users to register and execute workflows seamlessly by integrating with the REANA execution system. It also offers the capability to capture and visualize data provenance of the workflow executions, based on the W3C-PROV standard.
The project consists of three key components:
- FastAPI: Core component enabling RESTful API interactions with the platform.
- Keycloak: Authentication and access control system ensuring secure user authentication and user grouping
- MySQL Database: Database system serving as the backend for efficient data storage and retrieval.
All three components are deployed in a dockerized environment in order to ensure scalability, portability, and ease of management. This work assumes a platform and core components in line with the architecture specified as part of the HEurope project AI4Europe.
Key Features
- User authentication using keycloak
- Workflow registration (CWL workflows are currently supported).
- Integration with REANA system to execute previously registered workflows.
- CRUD operations both for registered and executed workflows.
- Data provenance for workflows executed within the REANA framework.
- Visualization of data provenance by generating graph-based PNG representations, allowing for clear and intuitive exploration of workflow dependencies and data flow.
Link: GitHub Repository
Severe Weather Detection
Description: Unsupervised Severe Weather Detection Via Joint Representation Learning Over Textual and Weather Data
Abstract: Official code repository of “Unsupervised Severe Weather Detection Via Joint Representation Learning Over Textual and Weather Data”. It includes the implementation of all experiments (using TensorFlow), as well as the scripts used to produce the figures displayed in the paper.
Link: GitHub Repository
Evidence Transfer
Description: Evidence transfer approach of combining categorical evidence to improve clustering tasks
Abstract: Official code repository of “Evidence Transfer for Improving Clustering Tasks Using External Categorical Evidence”. It includes the implementation of all experiments (using TensorFlow), as well as the scripts used to produce the figures displayed in the paper. It also includes all neural network configurations described using ANNETT-O. The ANNETT-O description can be found here.
Link: GitHub Repository
Presentations
Insights from the Dataset (Thodoris Aivalis)
Abstract: Generative Artificial Intelligence (AI) models have emerged as powerful tools across diverse applications. However, their inherent complexity often makes them opaque, which in turn complicates understanding their decision-making processes. Furthermore, it is often important to understand the effect training data have in the generation process without having internal view of the model. As an example, this is important for when using artistic or copyrighted material, or for awarding royalties, etc. In response to this challenge, we present a novel, search-inspired approach contributing to the interpretability of generative AI models by analysing the influence of training data on model’s outputs. Our method offers observational or external interpretability, meaning that it operates by observing the outputs of a model and not its internal activation patterns. Subsequently, we consider a range of similarity metrics based on both raw as well as latent-space embeddings produced by a pre-trained network. We evaluate the effectiveness of our method by retraining local models, demonstrating its ability to uncover influential subsets within the training dataset. Furthermore, this work lays the groundwork for future extensions and analyses, including user-based evaluations with domain experts such as artists and photographers. Validating the importance of the retrieved training samples in generated specific content more accurately, will contribute to improved observational interpretability.
Informed Machine Learning An Overview (Vassilis Gkatsis)
Preference Representation using Higher-Order LP (Antonis Trompoukis)
Neuro Symbolic AI (Antonis Ganios)
Abstract: In recent years, neural systems have demonstrated highly effective learning ability and superior perception intelligence. However, they have been found to lack effective reasoning and cognitive ability. On the other hand, symbolic systems exhibit exceptional cognitive intelligence but suffer from poor learning capabilities when compared to neural systems. Recognizing the advantages and disadvantages of both methodologies, an ideal solution emerges: combining neural systems and symbolic systems to create neural-symbolic learning systems that possess powerful perception and cognition. The purpose of this presentation is to illustrate different categories of neuro-symbolic models and frameworks related to them.