Which of the following approaches BEST helps to reduce model bias?

The correct answer is D. Ensuring diversity in training data sources. Model bias originates primarily from the training data. If training data overrepresents certain demographics, scenarios, or perspectives, the model learns and perpetuates those skews. Ensuring diversity in training data sources - spanning different populations, geographies…

AI Security Risk Management

Question

Options

AIncreasing the number of labels per instance
BDecreasing the frequency of model updates
CUtilizing a more complex model architecture
DEnsuring diversity in training data sources

How the community answered

(23 responses)

A
4% (1)
B
4% (1)
C
4% (1)
D
87% (20)

Explanation

Model bias originates primarily from the training data. If training data overrepresents certain demographics, scenarios, or perspectives, the model learns and perpetuates those skews. Ensuring diversity in training data sources - spanning different populations, geographies, contexts, and viewpoints - directly addresses the root cause of bias. Increasing labels (A) improves annotation granularity but doesn't fix underlying data skew. Decreasing model update frequency (B) has no bearing on bias. A more complex architecture (C) can actually amplify existing bias by fitting biased patterns more precisely.

Topics

#Model Bias#Data Diversity#Bias Mitigation#Training Data

Community Discussion

No community discussion yet for this question.

Full AAISM Practice