Research
Synthetic Images Match ImageNet in Vision Transformer Training
Formula-driven datasets may revolutionize AI by cutting costs and ethical issues while maintaining accuracy.
AI's Rapid Growth: New Frontiers and Challenges
AI's capabilities soar, but so do the challenges. From biocomputers to lunar data centers, what's next?
LVLM-VA: Aligning Vision Models with Human Insight
A new method enhances model robustness by reducing biases, aligning AI with human understanding.
AI Models as Students: Revolutionizing Learning Dynamics
A novel AI strategy empowers models to query for information, enhancing math and coding skills. Could this reshape educational AI?
Agent2World: Transforming Symbolic World Models with Multi-Agent Feedback
Agent2World's multi-agent framework redefines symbolic world models with interactive feedback and adaptive testing.
AI Models Embrace Student-Led Learning: A New Era of Interaction
AI models now improve by actively seeking information, promising advancements in education and healthcare efficiency.
Agent2World: Advancing World Models with Multi-Agent Feedback
Agent2World unveils a groundbreaking pipeline enhancing symbolic world models through dynamic multi-agent feedback.
New Method Accelerates AI Decision-Making by 142x
Innovative offline reinforcement learning approach enhances speed and reward efficiency in AI systems.
Bi-RAR: Elevating AI with Bidirectional Reasoning
Bi-RAR enhances AI accuracy in complex tasks by using bidirectional reasoning in retrieval-augmented generation.
fMRI-LM: Bridging Brain Imaging with Language Models
Introducing fMRI-LM, a breakthrough that links fMRI data with language models to enhance our understanding of neural activity.
ThinkingF: Advancing AI's Role in Mathematical Autoformalization
ThinkingF and StepFun-Formalizer-32B redefine benchmarks in converting natural-language math to formal expressions.
DeFloMat: Revolutionizing Clinical Object Detection with Speed and Precision
DeFloMat's Conditional Flow Matching dramatically cuts inference time, enhancing both speed and accuracy in medical imaging.
TableGPT-R1: Reinforcement Learning Elevates Tabular Data Analysis
TableGPT-R1 sets new benchmarks in tabular data reasoning, leveraging reinforcement learning to tackle data interpretation challenges.
ThinkingF: Breaking Barriers in Autoformalization with StepFun-Formalizer-32B
A new pipeline elevates AI's proficiency in translating math into formal language, setting unprecedented records.
Tiny-YOLOSAM: Revolutionizing Real-Time Image Segmentation
Tiny-YOLOSAM merges YOLOv12 and TinySAM for faster, efficient image segmentation, reshaping real-time computer vision.