ALGORITHM LAB.
ABOUT US
(1)
Current AI systems exhibit high performance through learning within a narrow range, but they also malfunction due to a lack of knowledge in other areas. Deep learning is limited to single-function roles such as image recognition. However, YOLO can recognize a considerable number of objects with a slightly lower recognition rate, all within a single system.
Language generation by LLMs synthesizes sentences from existing knowledge within existing text corpora, thus lacking the creativity to generate new sentences.
Image generation by GANs, following prompt instructions, generates similar images based on existing images or their seeds, allowing for the generation of various target images and even fake images.
紹介動画はこちら,video
Currently, research challenges include the inability to perform causal inference, integrate knowledge from different frameworks, and incorporate ethical concepts. While some AI agents under development have shown the ability to research previously unaddressed areas by combining existing research, examples of creative new general research are scarce. However, significant results have been achieved in limited areas such as molecular synthesis and pharmaceutical synthesis.
From these circumstances, in object recognition, due to the increasing number of types of objects to be recognized, it is necessary to improve the efficiency of hierarchical processing, such as coarse-grained image analysis, object limitation based on situational knowledge, and object limitation based on context. Overall, it is considered necessary to integrate and harmonize the operations of each AI, and to harmonize the hierarchical structure with lower-level AIs under higher-level concepts.
Furthermore, in LLM language generation, research on causal inference, alignment of results from different AIs, ethical concepts, interpretation and integration of laws, embodiment, automation of AI research, and scientific discovery are also important.
Algorithm_Lab. is promoting these research efforts and providing support to researchers. The International Conference on AI New Technologies and Open Discussions (ICAITD) aims to provide authors with useful and constructive feedback through the review process of academic presentations. During the period from acceptance to presentation, we strive to create an environment that fosters mutual understanding of research content and encourages further progress, ensuring fulfilling presentations and deepening confidence in the research.
ICAITD: https://www.icaitd.com