LLM-Driven AutoML for Cross-Lingual Handwritten OCR: Closed-Loop Neural Architecture Search with GPT-5, GPT-4o, and Claude Sonnet 4
Quick Answer
This study introduces a closed-loop AutoML framework utilizing GPT-5, GPT-4o, and Claude Sonnet 4 for cross-lingual handwritten OCR, achieving over 93% accuracy on Arabic, Persian, and English datasets without manual design.
Quick Take
The models demonstrate high efficiency with inference latencies between 41 and 44 milliseconds, showcasing the potential of as AutoML agents.
Key Points
- Framework autonomously generates and refines neural architectures using LLMs.
- Achieved mean test accuracy above 93%, with a peak of 98.1%.
- Evaluated through 270 independent experiments on diverse handwriting datasets.
- No manual architecture design or hyperparameter tuning required.
- Inference latency ranges from 41 to 44 milliseconds.
DeepSignal Analysis
What happened
A new AutoML framework has been developed that employs GPT-5, GPT-4o, and Claude Sonnet 4 for cross-lingual handwritten optical character recognition (OCR). This framework operates in a closed-loop manner, autonomously generating and refining neural network architectures. The models achieved over 93% accuracy on Arabic, Persian, and English datasets through extensive experimentation.
Key evidence
- The framework was evaluated through 270 independent experiments on Arabic, Persian, and English handwriting datasets.
- The generated models achieved mean test accuracies above 93%, with a peak accuracy of 98.1%.
- Inference latencies for the models ranged between 41 and 44 milliseconds, indicating high efficiency.
Why it matters
This study highlights the potential of large language models as effective agents in AutoML processes, particularly for tasks like handwriting recognition across multiple languages. The ability to achieve high accuracy without manual intervention could significantly streamline the development of OCR systems, making them more accessible and efficient. Furthermore, the closed-loop design allows for continuous improvement of model performance, which is crucial in dynamic applications.
Paper Resources
Source Excerpt
We present a fully automated closed-loop AutoML framework that uses GPT-5, GPT-4o, and Claude Sonnet 4 as autonomous neural architecture designers for cross-lingual handwritten optical character recognition. Each independently generates, trains, evaluates, and iteratively refines neural network architectures using performance feedback from previous trials. The framework is evaluated on Arabic, Persian, and English handwriting datasets through 270 independent experiments. It
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.