Warm-Start Multiplier Tuning for Machine Learning Fairness and Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models struggle to achieve high levels of fairness and accuracy simultaneously, with methods like random selection and retraining being inefficient and unpredictable, and existing optimization algorithms often require wasteful warm-up times due to random starting locations.
Innovation Solution
A warm start mechanism is introduced for a genetic algorithm that optimizes fairness and accuracy by using designed initial points, such as identity, parity, and opportunity points, to accelerate multiplier tuning and generate a tri-objective Pareto frontier for improved fairness and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If random selection or random generation is used to generate or adjust an inference, then exploration of the search space is enabled, but the inference becomes nondeterministic and may not be repeatable
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing lookup tables of adjusted inferences for different probability adjustments before runtime. This allows the system to perform deterministic lookups during inference adjustment rather than performing random operations, thereby maintaining repeatability while still enabling exploration through the pre-computed variety of adjustment options.
Solution Approach 2:
The patent creates copies of the base model's inferences through deterministic transformations using lookup tables. Instead of generating new random inferences, the system copies and transforms existing inferences through predefined adjustment mechanisms, ensuring repeatability while providing diverse exploration outcomes through multiple copy variants.
2Manufacturing precision
If preprocessing of training corpus or adjustment inside ML model is performed, then fairness and accuracy may be improved, but retraining is required which is slow
Solution Approach 1:
The patent extracts the fairness adjustment functionality from the model training process and relocates it to a post-processing stage. By taking out the adjustment logic from training and implementing it as a separate inference adjustment module with lookup tables, the system achieves fairness improvement without requiring slow retraining operations.
Solution Approach 2:
The patent introduces lookup tables as an intermediary between the base model's raw inferences and the final adjusted inferences. This intermediary structure enables fairness and accuracy improvements through deterministic transformations without requiring direct modification of the model parameters or retraining, thus avoiding time loss.
3Adaptability or versatility
If optimization algorithms use random starting locations, then exploration of search space is enabled, but much time is wasted identifying and evaluating many suboptimal solutions
Solution Approach 1:
The patent applies preliminary action by pre-computing lookup tables containing adjusted inferences for various probability adjustment scenarios before the optimization process begins. This eliminates the need for random starting locations and iterative exploration, as the system can directly access pre-computed optimal adjustments, dramatically improving optimization efficiency while maintaining comprehensive search coverage.
Solution Approach 2:
The patent skips the wasteful warm-up phase of random optimization by directly utilizing pre-computed lookup tables. Instead of rushing through many suboptimal random starting points, the system jumps directly to high-quality solutions through deterministic lookup operations, achieving the same exploration goal without the productivity loss.
Data Source
AI summary
Here is postprocessing calibration of class probabilities inferred by a machine learning (ML) model, and this calibration is improved by generation of novel initial points that accelerate a genetic algorithm that optimizes fairness and accuracy of the ML model. Tri-objective optimization for multiplier tuning is enhanced by adding a “warm start” mechanism. Innovative designed points are high performance as follows. An identity point has all group multipliers set to 1.0, corresponding to the original model. By definition, this solution will have high accuracy scores and no outcome regression. A parity point has multipliers that have near-perfect disparity and outcome regression. This entails finding multipliers that provide every subgroup approximately the outcome rate of the subgroup with the highest outcome rate. An opportunity point has multipliers that give an approximation of the outcome rate given by an Equality of Opportunity algorithm. This solution provides near-optimal values for accuracy and disparity.