How Can I Present My Deep Learning Thesis When Results Aren’t Improving?

At this point, making major changes to your results isn’t realistic, but that's okay! Focus on demonstrating a clear process in your thesis. Highlight the experiments you conducted, the different hyperparameters you experimented with, and discuss why certain strategies didn’t pan out. What really matters is that you showcase your investigative process. As for your advisor's feedback, it’s crucial to communicate the challenges you faced and ensure he supports your narrative. It's totally valid to write about what didn't work as long as you follow the scientific method.

Sounds frustrating! Just remember, your advisor's goal isn't to skew the data but to ensure that your findings and methods are solid. Don't let the pressure get to you. Emphasize what you did find out, the obstacles you encountered, and how they inform future work. Even 'negative' results can contribute valuable knowledge.

It's all about the journey, not just the destination! Make sure to document everything clearly—what you tried, what didn't work, and why. Discuss factors like dataset challenges, class imbalance, or potential limitations of the model. For your conclusion, frame your work as findings rather than failures. Talk about what you learned, acknowledge limitations, and suggest future avenues for research. It's not uncommon for theses with mixed results to pass if they’re well-documented.