Nota, a corporations specializing in AI model slimming and optimization, said on the 29th that its research to boost high-resolution image processing efficiency in vision-language models (VLMs) was accepted by ICLR 2026, the world's most prestigious AI conference.
ICLR is a conference where global big tech corporations such as Google and Meta, the parent of Facebook, showcase their flagship research. This year, the acceptance rate was only about 28%, reflecting a rigorous review.
According to Nota, the core of this research, the ERGO (Efficient Reasoning & Guided Observation) model, solves the massive computation expense issue that arises when AI processes high-resolution images. Unlike existing models that rely on "perception-centric reasoning" that mechanically sweeps entire images and thus miss information at low resolution, ERGO introduces a "reasoning-based perception" approach that uses multimodal context to decide on its own where to focus.
The ERGO model first grasps the overall scene with a low-resolution image, then selectively zooms in on only those regions with visual uncertainty or requiring fine-grained analysis to examine them at the original resolution. The company said it applied a reward system developed through a Reinforcement Learning framework, giving the model the intelligence to autonomously find the most suitable region to answer a question.
In actual benchmark tests, ERGO used only 23% of the visual tokens compared with state-of-the-art AI models while delivering high accuracy. By maximizing computational efficiency, its inference speed was about three times faster than previous approaches.
Nota applied the technology to the Nota Vision Agent (NVA). A company official said, "NVA is a real-time video monitoring solution based on VLMs, and we expect this research to further accelerate adoption in the physical AI field, where instantaneous judgment is essential." The strength of this technology is that it achieves high-resolution, fine-grained inference using only low-spec edge devices, without expensive servers.
Based on this, Nota plans to deploy ERGO technology across global projects, including the UAE Intelligent Transportation System (ITS) initiative currently underway.
Chae Myeong-su, CEO of Nota, said, "This ICLR acceptance means Nota's AI slimming and optimization technology has moved beyond merely reducing model size to the stage of 'intelligent optimization' that streamlines how AI thinks," adding, "We will apply the validated ERGO technology to NVA to deliver expense-efficient and accurate visual intelligence across various industrial sites."