Jiani Hu

Title: Jiani Hu: Inventor in Object Recognition and Neural Network Training.

Introduction

Jiani Hu is an inventor based in Beijing, China, whose patent work focuses on training neural networks for object recognition. Hu is associated with Canon Kabushiki Kaisha and has 3 patents.

Latest Patents

Hu’s listed patent work includes methods and apparatus for training object-recognition models. One invention describes optimizing training samples by measuring fluctuations in a model’s predictions relative to the samples’ labeled identities. The apparatus uses those fluctuations to determine whether each sample should be included in the next training epoch.

Another invention concerns training a neural network using simple and hard image samples. The method transfers features between sample types, calculates distillation and classification losses, combines them into a total loss, and updates the network’s parameters.

Career Highlights

Hu’s patent activity reflects work on improving neural-network training processes for object recognition. The described approaches address sample selection, feature transfer, and model optimization.

Collaborations

Hu’s listed coworkers include Weihong Deng and Xian Li. Hu is associated with Canon Kabushiki Kaisha, a Japanese company known internationally as Canon.

Conclusion

Jiani Hu’s inventions focus on practical methods for training object-recognition neural networks, including optimizing training samples and combining feature distillation with classification.