Preferred Networks
Japanese deep learning research and technology company
Preferred Networks is a Japanese company focused on deep learning research and its application to industrial problems, including robotics, manufacturing, drug discovery, and autonomous driving. Originally known for developing the Chainer…
Definition
Preferred Networks is a Japanese company focused on deep learning research and its application to industrial problems, including robotics, manufacturing, drug discovery, and autonomous driving. Originally known for developing the Chainer deep learning framework, it works closely with major Japanese industrial partners such as Toyota and Fanuc to apply machine learning to real-world manufacturing, mobility, and materials science problems across heavy industry.
Overview
Preferred Networks was founded on the premise that deep learning's greatest near-term value in Japan would come from applying it to the country's strong industrial and manufacturing base, rather than pursuing consumer internet applications as many Western AI companies had. This orientation shaped the company from the outset toward close partnerships with established industrial firms rather than building consumer-facing products, positioning it as a technology supplier to heavy industry as much as an AI research company in its own right. Mechanically, Preferred Networks operates as both a research organization, having created the Chainer deep learning framework that influenced the design of subsequent tools before the team shifted focus toward frameworks like PyTorch, and an applied engineering company delivering AI systems to industrial partners across robotics control, image-based quality inspection, drug discovery simulation, and materials science. This mix requires the company to maintain both cutting-edge research capability and the practical engineering discipline needed to deploy AI systems in industrial and manufacturing environments, which have much stricter reliability and safety requirements than typical consumer software. Among AI companies, Preferred Networks is distinguished from both large-model-focused labs like OpenAI or Anthropic and from academic labs like MIT CSAIL by its deep specialization in applying deep learning to physical, industrial systems rather than to language or general-purpose reasoning. It occupies a niche closer to applied industrial AI and robotics companies than to foundation model developers, with major backing and collaboration from Japanese manufacturing and automotive giants rather than internet or software companies. In practice, Preferred Networks' work has appeared in projects with Toyota on autonomous driving and mobility research, with Fanuc on industrial robotics and factory automation, and in materials science and drug discovery collaborations that apply machine learning to accelerate simulation and design processes traditionally reliant on slower physical experimentation. The company has also historically served as a prominent example of Japan's own AI research capability distinct from the US and Chinese AI ecosystems. Because Preferred Networks' business model depends heavily on deep, long-term partnerships with a relatively small number of major industrial clients, its trajectory is closely tied to those relationships and to the industrial sectors it serves, rather than to a broad consumer or developer user base. Its early framework, Chainer, was eventually deprecated in favor of the team contributing to and adopting other tools, illustrating the trade-off smaller research organizations face in trying to sustain their own deep learning framework against better-resourced competitors like Meta's PyTorch or Google's TensorFlow.
Key Concepts
- Japanese company applying deep learning to industrial problems
- Created the Chainer deep learning framework
- Partners closely with Toyota on autonomous driving research
- Collaborates with Fanuc on industrial robotics and automation
- Applies machine learning to drug discovery and materials science
- Focuses on physical and industrial AI rather than consumer products
- Combines research capability with industrial engineering discipline
Use Cases
Frequently Asked Questions
From the Blog
Neural Networks Explained with Simple Analogies
Neural networks are webs of simple units trained to recognise patterns — explained here without any maths.
Read More AI & TechnologyWhat Is Self-Attention in Neural Networks
Self-attention lets each token in a sequence attend to every other token in the same sequence, building context-aware representations that power transformer models.
Read More AI & TechnologyWhat Is Backpropagation in Neural Networks
Backpropagation is the algorithm that lets neural networks learn by efficiently calculating how much each weight contributed to the error and adjusting it.
Read More AI & TechnologyWhat Are Activation Functions in Neural Networks
Activation functions add non-linearity to neural networks, letting them learn complex patterns instead of behaving like a simple linear model. Here's how they work.
Read More