Benefits
We believe the project can bring various business values to our primary stakeholders.
Specifcially, the project can benefit the Xbox team, as well as the Xbox users.
To Engineers
The deliverables of our project have provided pipelines for similar algorithms evaluation, which the engineers in the Xbox team can easily extend and apply in future work to compare more target algorithms
To PMs
Our work also helps the program managers in the Xbox team decide between the current two algorithms on which algorithm to use in given contexts. Moreover, we built the data analysis pipeline for the data team to generate visual reports. These benefits of decision-making will potentially increase the efficiency and effectiveness of the program management work in the Xbox team.
To Customers
With more understandings of the image detection algorithms, the Xbox team will make progress on detecting and filter the unsafe contents in the game images more effectively. This will improve the users' experience, increase customer satisfaction, and further help to deliver better services to the Xbox users.
Future Work
Next steps on both the engineering and data science aspects we suggest our sponsor to focus on
- All
- Engineering
- Data Science
Comparison Options
New options for comparison can be added to just compare the base image to certain chosen test image sets
Integration
Can use interfaces to integrate the current two algorithms and even more target algorithms in future work
Use of Cache
The results files that the current system can generate might not be small enough. A cache can be used in the future to downsize the files.
Parallel Processing
The current parallel processing system can be further adjusted to adapt to the future needs of larger datasets.
Stability Improvement
The stability of the system that obtains results from the second algorithm (Tech B) can be further improved.
Statistics Hypothesis
When we have more sample data, we can use hypotheses and tests to see if the sample data can represent the population or not.
Automated Comparison
Use an ML model and the train-test split to evaluate the performance of technologies. It is also beneficial to apply to more algorithms.
Higher Dimensions
Include more features to improve the performance of each algorithm.