LOS ANGELES, April 02, 2026 (GLOBE NEWSWIRE) — Fotor, the flagship AI product owned by Everimaging, today announced that its latest joint research result has been accepted as a conference paper at ICLR 2026, one of the world’s leading academic conferences in artificial intelligence.
The paper, titled “WEB-COGREASONER: TOWARDS MULTIMODAL KNOWLEDGE-INDUCED COGNITIVE REASONING FOR WEB AGENTS,” represents a significant milestone in the era of autonomous AI operations. This breakthrough was achieved through a collaborative effort by Fotor and prestigious universities/research institutions, successfully bridging the gap between academic achievements and real-world industrial practice.
Core Breakthrough: Letting AI Master “Triple Knowledge”
Facing complex webpages and pop-up interference, traditional web agents often fall into dead loops. To break this bottleneck, the joint research team drew inspiration from “Bloom’s Taxonomy” in educational psychology and innovatively proposed the Web-CogKnowledge framework.
This framework deconstructs the AI learning process into three progressive cognitive stages:
- Factual Knowledge: Enabling the AI to accurately identify web elements and predict the direct consequences of single interactions.
- Conceptual Knowledge: Granting the AI deep semantic understanding to infer component functions and grasp overall webpage intent.
- Procedural Knowledge: Training the AI in planning and decision-making to execute action sequences and handle unexpected interruptions like pop-ups.
Supported by this system and the Web-CogDataset (constructed from 14 real-world websites), AI now possesses a powerful “Knowledge-driven Chain of Thought (CoT)” for deep logical reasoning.
From Research to Product: Powering the Next Generation of Fotor Agent
Building on its ICLR 2026 research, Fotor has been equipped with the capability of translating academic breakthroughs into product excellence by integrating these advances with leading open-source agent frameworks (e.g. OpenClaw), making the next evolution of Fotor Agent within reach.
Under the Web-CogReasoner framework, Fotor Agent is moving beyond dependence on a webpage’s underlying code (DOM tree). By leveraging “pure pixel visual perception” and the cross-platform cognitive framework presented in the joint paper, Fotor Agent can seamlessly manage workflows across web versions, desktop software, and mobile apps. This marks a definitive step toward Universal Computer Control (UCC), allowing users to handle complex tasks through a single command.
About Fotor
Dedicated to the practical application of top-tier AI research, Fotor bridges the gap between complex neural architectures and intuitive creative tools. Through its robust global research network, Fotor continues to bring new AI experiences to more than 800 million users worldwide.
Media Contact:
CONTACT: Coran Huang
EMAIL: coran@fotor.com
WEB: http://www.fotor.com
A photo accompanying this announcement is available at https://www.globenewswire.com/NewsRoom/AttachmentNg/ecf0389c-bd6e-4585-a07b-f307bf915d7e
