Lukas Enderich

Title: Lukas Enderich and Fast Quantised Training

Introduction

Lukas Enderich is an inventor associated with Stuttgart, Germany, and Robert Bosch. His work includes a patent for a method designed to make training trainable modules faster while accounting for quantised parameter values.

Latest Patents

Enderich is associated with 1 patent, titled “Fast quantised training of trainable modules.” The invention concerns training a module that maps input variables to output variables through an internal processing chain. It uses learning data containing input values and their corresponding output values, along with a list of discrete parameter values selected for lossless storage.

During training, the module maps the inputs to assessment outputs. A cost function measures both the difference between the assessment outputs and the learning outputs and the difference between at least one internal parameter and a discrete value. At least one parameter is then adjusted to improve the cost function. This approach incorporates quantisation into the training process.

Career Highlights

The available information identifies Enderich as working at Robert Bosch. His patent focuses on training methods that account for discrete, efficiently stored parameter values.

Collaborations

Fabian Timm is identified as Enderich’s coworker. The provided information does not specify whether Timm collaborated on the patent.

Conclusion

Lukas Enderich’s patent describes a quantisation-aware training method that aims to improve trainable modules while keeping their parameters compatible with lossless storage.