| Budget Amount *help |
¥25,870,000 (Direct Cost: ¥19,900,000、Indirect Cost: ¥5,970,000)
Fiscal Year 2024: ¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000)
Fiscal Year 2023: ¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000)
Fiscal Year 2022: ¥7,410,000 (Direct Cost: ¥5,700,000、Indirect Cost: ¥1,710,000)
Fiscal Year 2021: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Fiscal Year 2020: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
|
|---|
| Outline of Final Research Achievements |
We devised and validated a method to objectively evaluate how closely generative models, obtained through advanced scientific and technological means, approximate human capabilities from the perspective of power laws. By verifying the well-known power law in natural data, such as natural language, financial data, and image data (especially medical data), we demonstrated that the performance of generative models can be assessed based on the data generated by these models. In natural language, we showed that the power law can evaluate the performance of language models, which have seen remarkable advancements in recent years. A book summarizing this research received the Mainichi Publishing Award. Furthermore, we explored an untapped new approach within power laws and showed its potential to contribute to the evaluation of future advanced generative models.
|
