Participation at IUI 2025
On March 24-27, 2025, the e-Lite group will attend (in person) the 30th International Conference on Intelligent User Interfaces (IUI 2025) in Cagliari, Italy.
Tommaso Calò will present the paper "DeepFlow: A Flow-Based Visual Programming Tool for Deep Learning Development" on Thursday, March 27 during the Democratisation of AI session at 10:05 AM.
The paper introduces DeepFlow, a visual tool that makes deep learning development more accessible by allowing users to build models by connecting visual blocks rather than writing code. DeepFlow solves two major challenges: it helps users create complex, large-scale models through a feature called "supernodes" (groups of elements that can be reused), and it provides immediate feedback about errors before models are run. This visual approach makes it easier to understand how different parts of deep learning models connect and work together, similar to how flowcharts help visualize complex processes.
A user study with 16 deep learning developers explored how people with different levels of expertise used DeepFlow for typical model design tasks. Results revealed DeepFlow's dual value: novices found it served as an effective learning aid for understanding complex architectures, while experts appreciated its capabilities for rapid prototyping of models.
This work contributes to our ongoing effort to democratize deep learning tools, making them more accessible while maintaining the capabilities needed for advanced applications.
Additional information: