commit.

2025-06-05 10:18:13 +02:00
parent 9b881bf058
commit b2116c22df
6304 changed files with 464 additions and 708461 deletions
@@ -1,49 +0,0 @@
---
-tags: 
- "#modeling" 
- "#machinelearning"
- "#personalthoughts"
- "#ideas"
- "#readingnotes"
---
-# AI and Modeling
-AI is not a threat for modelers, since our job is both creative and requires empathy and human interaction. It seems there is a list of [[top 10 safest jobs from AI]]. Modeling can take advantage of AI and also give place to some relevant contributions in the ML field. In particular, 
-
- **modeling can take advantage of techniques and tools from the AI community** by means of recommendation systems (or named model assistance tools) which can support modelers during their tasks. For instance, take a look at the [model-based bot development framework](https://modeling-languages.com/multi-platform-chatbot-modeling-deployment-jarvis/). In general we could try to develop better and  [smart modeling tools](https://modeling-languages.com/smart-modeling-tools-ai-software/);
- **the AI community can be helped by the modeling community** to improve for instance the specification of ML pipelines ([modeling Machine Learning pipelines](https://modeling-languages.com/tools-modeling-artificial-intelligence-code/) (built by ML experts but not modeling experts, we can do much better!!)); to introduce architectures and software engineering practices in general to make better tools and enhance the reusability of ML components. Currently, they are assembled in a bespoke manner. 
-
-
-# Ideas
- **Check the appropriateness of our (meta)model encoding mechanisms (based on uni/bi/multi-gram) vs the adoption of [graph kernels)](https://modeling-languages.com/graph-kernels-model-driven-software-engineering/**
- **Recommendation systems for supporting the development of models**
-	- The main idea is to use collaborative filtering techniques for recommending model items
-	- We need to conceive some mechanisms to configure the features to be represented in the user-item matrix and how
-	- Maybe the configurations depend on the corresponding metamodels
- **Recommendation systems for supporting the development of model-to-model transformations**
-	- We need to decide which technology we want to focus first (ATL, ETL, etc.)
-	- It would be interesting to conceive something similar to what we have done with FOCUS for recommending software developers with API function calls and usage patterns
- **Recommendation systems for supporting the development of model-to-text transformations**
-	- We can focus on Acceleo
-	- As for the previous point, it would be interesting to provide something similar to FOCUS
- **Recommending Stackoverflow posts that are relevant to the modeling activity being performed**
-	- We would like to do something similar we have done with PostFinder to recommend Java developers with posts that are relevant to the current development tas2k
-	- Main challenges:
-		- Make the approach done in PostFinder generic. In particular all the steps like code wrapping etc. have to be mapped in the modeling domain. For instance, concerning the code wrapping phase, can we rely on atlanalyser? What happens if a parser is not available?
- **A low-code approach to support the development of recommendation systems for modelers**
-	- It can be a feature model based approach
-		- In the Extremo paper, authors identified the feature models related to the functionalities provided by modeling assistants to the final users. In this work, we would like to explore the functionalities that might be configured for developing a modeling assistant. Here, the focus would be the developer of the modelling assistant and not its final users (covered by the Extremo paper) 
- **Recommending similar MDE artifacts by means of the MNBN approach**
-	- The idea is to reuse the work done for our EASE paper to recommend similar artifacts by using the MNBN 
-	- The inputs should be textual files (model, M2M transformation, meta-model) to be classified 
-	- Main challenges
-		- Check if the TF-IDF Vectorize is suitable for this domain 
-		- Find a labelled dataset to train the network 
-		- Recommending engine to be integrated with the network
-	- Initial results on ATL transformations
- Joint paper about a **research agenda presenting what are the goals, challenges, etc, around the topic of recommendation systems for MDE**. This can be a good paper for a models workshop, e.g., MDE and AI. What do you think? The structure of the paper can reflect the topics we have identified in the list above. 
-	- All the recommendation systems previously identified can be part of a research agenda that we can motivate by referring to what currently happens in software development (here we also have all the work done in CROSSMINER).
-	- For each recommendation system we could list the needed ingredients and mainly the existing issues that might currently hamper their realization.
-
-
-
-