15/07/2026
At one of the world’s most recognized and authoritative conferences in artificial intelligence, ICML 2026, STATANLY Technologies presented a series of research works covering AI agents, large language models, computer vision, time-series forecasting, vector graphics, scientific video generation, and industrial digital twins.
The presented works include:
GENEGO: A Disciplined LLM Negotiation Agent with Pairwise Preference Inference
A negotiation agent based on large language models that infers preferences through pairwise comparisons and helps structure negotiations in a more transparent and disciplined way.
SVG-BERT: Code-Native Representations for Structure-Aware Vector Graphics Workflows
A method for working with SVG not just as an image, but as structured code. This enables AI systems to understand objects, layers, relationships, and the internal logic of vector graphics.
Object-Level Access to SVG Scenes
A benchmark and method for extracting individual vector objects from SVG scenes using text guidance and raster-based grounding. This is relevant for design automation, visual editing, and generative interfaces.
Blind Frames, Broken Stories: the Verification Bottleneck in Scientific Video Generation
A study of verification challenges in scientific video generation, where a video may look realistic but still break physical, scientific, or causal logic.
Forecasting Time-Varying Correlation Matrices with Large Language Models
Research on using large language models to forecast changing relationships between variables over time — an important task for complex systems, finance, risk analysis, and industrial analytics.
TARPA: Retrieval Priors for Time-Series Foundation Models on Audience Retention
A time-series forecasting approach that uses retrieval-based priors — similar historical patterns — to improve audience retention prediction.
Digital Twin Builder: A Multi-Agent LLM System for Automated Industrial Digital Twin Development
A multi-agent system that helps automate the creation of industrial digital twins — virtual models of production systems, assets, or processes used for analysis, simulation, and optimization.
For STATANLY Technologies, these works reflect our broader direction: moving from isolated AI models to reliable, verifiable, and production-ready AI systems that can be applied in real business and industrial environments.
We are also proud that STATANLY Technologies was among the Russian companies presenting research at ICML 2026, alongside Yandex, Sber, and T-Technologies.
