Projects funded under national AI programme AISG focus on good training datasets

Using only visual photos is insufficient for accurately sorting plastic waste, as machines struggle to distinguish between different material types without deeper data. By incorporating hyperspectral imaging, which measures material reflectivity to identify molecular structures, AI systems can achieve high accuracy in categorizing plastics. This approach solves a critical bottleneck in recycling infrastructure, where conventional sorting methods fail to separate materials effectively, thereby preventing contamination and improving recycling rates. The project highlights that the primary barrier to effective AI adoption is often a lack of high-quality, specialized training data rather than computational power alone. Successful implementation requires specific datasets that enable machines to perceive information beyond human visual capabilities, such as extended wavelengths. Without this refined data, AI models remain inaccurate and unreliable, underscoring that data curation and relevance are more vital than simply increasing the volume of available information. This development is highly relevant to open data because the underlying principle—that domain-specific, high-fidelity data drives trustworthy AI—applies across all sectors. As organizations strive for transparency and efficiency, they must prioritize the creation and sharing of robust, well-labeled datasets. Making such specialized data openly available would accelerate innovation, reduce redundant training efforts, and ensure that AI solutions are reliable enough to be widely trusted and adopted.

Source: straitstimes.com
Published on 2024-07-08

Milei contrató seis encuestas por $ 52 millones y analiza el clima en medio del ajuste »

Open Data

Related news