MemoraX AI at NeurIPS 2026: Ten Papers on Reliable Learning, Efficient Reasoning, and Long-Term Agent Memory
We are excited to share that 10 papers from MemoraX AI have been accepted to NeurIPS 2026 . These papers cover a range of topics, including robust self-training, reinforcement learning, difficulty-aware reasoning, diffusion models, long-term Agent Memory, and the continuous evolution of AI systems. Although the papers address different technical problems, they share a common research question:…
MemoraX AI has had ten papers accepted for presentation at NeurIPS 2026. These papers explore various aspects of reliable learning, efficient reasoning, and long-term agent memory in AI systems. Rather than focusing on larger models, the research emphasizes recognizing when a problem requires deeper reasoning, learning from reliable experience, and retaining useful information over time.
One line of work addresses improving the reliability of self-training for large language models. The researchers propose a robustness-driven evolutionary self-training framework that ensures solutions are not only correct but also robust, stable, and transferable to related problems. Another research direction examines the gap between training improvements and real inference behavior.
A framework is introduced to optimize inference policies rather than just training policies, allowing potentially harmful updates to be detected and rejected.
The research also investigates ways to make reasoning more compute-efficient. A difficulty-perception mechanism is proposed to help models dynamically adjust reasoning intensity based on problem difficulty. This can improve token efficiency in practical AI systems. Additionally, diffusion models are explored for online reinforcement learning, with a proposed approach called Coupled Flow that balances expressiveness with optimization efficiency.
A central research area for MemoraX AI is long-term agent memory. The company's Coding Agents, such as Codex and Claude Code, are designed to retain and reuse project context, historical decisions, debugging experience, and reusable development workflows. MemoraX Code aims to help these agents maintain continuity across long-running development tasks.
Finally, the research also focuses on evaluating Agent Memory systems. MemoraX AI is developing a reproducible evaluation framework for comparing memory systems across various scenarios, such as long conversations, coding memory, multimodal memory, and temporal understanding. This will help clarify the capabilities and limitations of different memory systems in AI agents.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
