A research team is developing a generative AI model to create novel protein structures based on a sequence of amino acids. They want the generated structures to maintain realistic physical properties and chemical bonds. While the model can generate diverse structures, it sometimes produces outputs that are chemically impossible or unstable. What core concept of generative AI is the team struggling to ensure in their model's output?
- AControllability
- BDiversity
- CNovelty
- DFidelity
Show answer & explanationAnswer & explanation
Correct answer: D. Fidelity
The team is struggling with 'Fidelity'. Fidelity in generative AI refers to the quality of the generated output being realistic, accurate, and adhering to the underlying data distribution and constraints (e.g., realistic physical properties and chemical bonds for protein structures). While the model might achieve diversity and novelty, its lack of fidelity means the generated outputs are not high-quality or 'real' enough. Controllability refers to guiding the generation process.
Why the other options are wrong
- A. Controllability refers to the ability to guide or influence the characteristics of the generated output, which is not the primary issue described.
- B. Diversity refers to the variety of outputs a generative model can produce, which the team seems to achieve ('diverse structures').
- C. Novelty refers to the model's ability to create unique outputs that are not simply copies of the training data, which is implied by 'novel protein structures'.
Fidelity (Generative AI)
A key evaluation criterion for generative AI models, referring to the quality, realism, and accuracy of the generated outputs, and their adherence to the underlying data distribution and real-world constraints.
- Measures how 'real' or high-quality the generated content is.
- Crucial for applications where outputs must be physically or semantically correct.
- Often balanced against diversity and novelty in model design.
Memory trick: Generative AI needs Fidelity, Diversity, and Novelty.