When developing an algorithm under tight deadlines, I think the first step is ensuring you’re on the right path. Quick POCs help validate the riskiest or most uncertain aspects early. Context is key—how much risk can you take? Does the domain allow failures and incremental improvements, or do you need solid results now? As for cutting-edge vs. proven methods, I believe it’s crucial to stay practical. Cutting-edge is tempting but should only be chosen if it offers clear advantages. Otherwise, proven methods are my go-to, especially when time is tight. It’s all about balancing ambition with pragmatism.

The optimal choice is to prioritize methods that balance familiarity with the specific requirements of the task, recognizing that "cutting-edge" and "proven" are not mutually exclusive. By focusing on approaches that the team is adept at implementing, regardless of their novelty, you ensure both reliability and efficiency under tight deadlines.

I would played it safe! Developing a new algorithm needs time in a relaxed environment. If you have the resources you can go with cutting edge algorithms but always have a backup algorithm just in case you need it.

1. Reliability vs. Risk • Proven methods offer reliability and have been extensively tested, reducing the chance of failure. • Cutting-edge methods may yield superior performance but carry higher risks due to limited testing and documentation. 2. Time Constraints • Tight deadlines favor proven methods because they allow faster implementation with fewer unexpected challenges. • Cutting-edge methods may involve steep learning curves and experimentation, which can delay progress. 3. Team Expertise • A team experienced with proven techniques can deliver results efficiently. • For cutting-edge methods, prior knowledge or experience is critical; otherwise, they can lead to bottlenecks.

I think it’s good to proceed with proven methods first if the product needs to be delivered within tight deadlines because with cutting edge technologies it’s a known risk that it might not work that well on our specific use-case for example, unet might give better performance for the task of image segmentation than segformer.

Depends on the priority! Developing a new algorithm is time consuming where multiple corner cases should be looked upon albeit at the same time if clock is ticking then cutting-edge methods can be helpful. Trade-off.

Under tight deadlines, the priority should be on delivering the product efficiently by relying on proven methods rather than experimenting with new approaches.

Hybrid Approach: Combine proven methods for reliability with cutting-edge techniques for future improvement. MVP First: Build a minimum viable product using proven solutions, then iterate with innovative methods. Time-Boxed Experimentation: Allocate time to test new methods; revert to proven solutions if unsuccessful. Parallel Exploration: Divide the team to work on both proven and cutting-edge methods simultaneously. Leverage Community Tools: Use open-source libraries and tools to accelerate development and integrate new techniques.

Here comes the agile methodology, start with proven one and gradually move/migrate to cutting-edge or hybrid one. It will also depend on resource availability and quality of resources available.