Advanced Computing in the Age of AI | Wednesday, May 22, 2024

NeoML Released as TensorFlow Alternative 

A new open source library for training machine learning models is billed as rivaling the performance of AI models trained with established libraries like TensorFlow, especially models running on a mobile device.

NeoML, described as a “cross-platform machine learning framework” supporting deep learning and ML algorithms, was launched this week on GitHub by ABBYY. The Bay Area enterprise software developer specializes in applications like optical character recognition and linguistic software. ABBYY said it currently uses NeoML for computer vision and natural language processing applications, including image processing and data extraction.

Running in the cloud, desktops or on mobile devices, NeoML is touted as running up to 20 times faster than TensorFlow for pre-trained image processing models, regardless of device. “The combination of higher inference speed with platform-independence makes the library ideal for mobile solutions that require both seamless customer experience and on-device data processing,” the company claimed.

The new ML library is being promoted for deploying object identification, classification and semantic segmentation as well as predictive modeling. Backers also claimed NeoML makes more efficient use of available cloud resources for retail applications like tracking consumer preferences.

NeoML supports the Open Neural Network Exchange (ONNX) standard backed by Facebook (NASDAQ: FB), Microsoft (NASDAQ: MSFT)

and others. The open ecosystem for interoperable machine learning models is designed to improve tool compatibility.

NeoML’s single code base can run across operating systems from Linux and Windows to Android and iOS. The library also is optimized to run on both CPUs and GPUs. It currently supports the C++, Java and Objective-C programming languages, with Python to be added soon.

The open source release includes more than 20 “traditional” machine learning algorithms for classification, regression prediction models and clustering, ABBYY said.

Speed comparisons between NeoML and TensorFlow were conducted on x86, Nvidia CUDA-enabled graphics and Arm-64 processors running on both Android and iOS devices. According to internal test results released by ABBYY, the most recent version of NeoML outperformed the TFLite 2.1.0 version of TensorFlow in most scenarios.

Along with faster inference, Ivan Yamshchikov, ABBYY’s AI evangelist, also stressed NeoML multi-platform capabilities, “especially its potential on mobile devices.”

About the author: George Leopold

George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).