https://conf.researchr.org/track/fse-2025/fse-2025-ideas-visions-and-reflections
- Visions of the future, such as:
- Bold visions of new directions that may not yet be supported by solid results but rather by a strong and well-motivated scientific intuition. Examples include unusual synergies with other disciplines, or the importance of software engineering in problems whose software engineering aspects have not been studied earlier.
- Summaries of highly innovative research ideas recently awarded as grants.
Write a vision paper on energy efficient AI pipeline from end to end, both horizontal and vertical dimension. The ACM International Conference on the Foundations of Software Engineering (FSE) is the target.
data compression, memory optimization, aiops, activation func, data transfer, storage, reduce build failures, horizontal and vertical, right from datacollections to model inference, recycling, reuse, or if not then reduce time/compute/memory, smaller models, concious of what needs to be done and shouldn’t be an overkill, energy is not just the energy for running model but it starts right from the data collection, human efforts, annotation, take motivation from a different domain e.g., economics.
Optimal path solution, state automata, cost for each path, each step has multiple combinations or options to choose from, we need to chose the most optimal solution and strategy thats in sync with current, past and future state.
Vision for energy efficient llm supprted software engineering - Goal is to get apprpriate if not better shaped software output, compared to only if human worked, and ensuring we save energy, this will entail time it took to develop earlier vs now. Can we compare time it took for tasks earlier before LLM inception vs now, maybe compare story completion time.
In my vision paper, I want to introduce a theoretical framework that give a vision of energy efficiency of AI in context of software engineering. The theoritical framework would help researchers and developers save energy in context of both model development, model usage, development practices, etc, basically all aspects involved between any interaction between developers and models.
Now i want help designing this theoritical framework. Since there are various types of interaction between developers and AI models, and each of such interaction has different ways to conduct those interaction each having different energy cost involved, which add’s up, and at the end of the interaction, which may involve different stages, choosing one option or possibility can affect the final overall energy consumption, and we want to ensure the energy consumption is optimal. I want the theoritical framework to have real world usage and implication, based on this information, What principle, laws etc., can be borrowed or used from any other domain in the world, in this context of optimization of energy for software engineering ai enigneering pipeline/workflow
We are trying to automate every step in software development, will we have a future where ai writes code or humans tell ai to write code, or ai tells humans to write the code. One factor is how good the step is, which we are seeing it’s gettinf better. However this replacement has energy concerns, as running ai models take energy. We need a plan as software development community to ensure, we move forward in a sustainable way that paves way for a greener ai usage in software engineering. We need to develop and adapt software development lifecycle stages to incorporate this. This vision paper paints a picture of how a sustainable future of our software development community should look like.
Nov 24
Currently AI developement pipeline is