Machine Learning
Interpretation
Weiyuan @ SFU DSL
Machine learning becomes popular in lots of domains.
- Learned Index
Machine learning becomes popular in lots of domains.
\(\gamma((R \bowtie S) \bowtie T\)
\(\gamma(R \times S) \bowtie T\)
VS
- Learned Index
- Learned Query Plan
Machine learning becomes popular in lots of domains.
\(\gamma((R \bowtie S) \bowtie T\)
\(\gamma(R \times S) \bowtie T\)
VS
- Learned Index
- Learned Query Plan
Postgres Plan: 18.3s
Learned Plan: 3.9s
Machine learning becomes popular in lots of domains.
\(\gamma((R \bowtie S) \bowtie T\)
\(\gamma(R \times S) \bowtie T\)
VS
- Learned Index
- Learned Query Plan
Avoid Cartesian Products is a Common Heuristic!
Postgres Plan: 18.3s
Learned Plan: 3.9s
Machine learning becomes popular in lots of domains.
- Learned Index
- Learned Query Plan
- Learned DB Parameter
Machine learning becomes popular in lots of domains.
- Learned Index
- Learned Query Plan
- Learned DB Parameter
- ER
- ....
EU General Data Protection Regulation
“shall have the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her“.
Not Understand & Important
Interpretation
Because of
- Safety: We want to make sure the system is making sound decisions.
- Debugging: We want to understand why a system doesn't work, so we can fix it.
- Science: We want to understand something new.
- Legal/Ethics: We're legally required to provide an explanation and/or we don't want to discriminate against particular groups.
- ...
- Case-based
- Feature-based
- Model-based
Types of interpretations
- Case-based
- Feature-based
- Model-based
Types of interpretations
- Case-based
- Feature-based
- Model-based
Types of interpretations
- Safety: We want to make sure the system is making sound decisions.
- Debugging: We want to understand why a system doesn't work, so we can fix it.
- Science: We want to understand something new.
- Legal/Ethics: We're legally required to provide an explanation and/or we don't want to discriminate against particular groups.
- ...
- If there is an issue with the data
- If so, where the issues are
- What the issue is
- How to fix.