Tong Meng

Title: Innovations of Inventor Tong Meng

Introduction: Tong Meng is a prominent inventor based in Beijing, China. He has made significant contributions to the field of deep neural networks, particularly in the area of fixed-point training methods. With a total of 2 patents to his name, his work is paving the way for advancements in artificial intelligence and machine learning.

Latest Patents: Tong Meng's latest patents include two innovative methods for training deep neural networks. The first patent, titled "Fixed-point training method for deep neural networks based on static fixed-point conversion scheme," proposes a method and apparatus for LSTM neural networks. This method utilizes fixed-point numbers during the fine-tuning process, allowing the network to achieve desired accuracy levels under floating-point calculations within several training cycles. The second patent, "Fixed-point training method for deep neural networks based on dynamic fixed-point conversion scheme," similarly focuses on LSTM neural networks. It also employs fixed-point numbers for forward calculations during fine-tuning, ensuring that the network accuracy can be restored to the desired level.

Career Highlights: Tong Meng is currently employed at Xilinx Technology Beijing Limited, where he continues to develop innovative solutions in the field of neural networks. His expertise in fixed-point training methods has positioned him as a valuable asset in the technology sector.

Collaborations: Tong Meng has collaborated with notable colleagues, including Xin Li and Song Han, who share his passion for advancing neural network technologies.

Conclusion: Tong Meng's contributions to the field of deep neural networks through his innovative patents highlight his role as a leading inventor in the technology landscape. His work continues to influence the development of artificial intelligence applications.