Training LBNsLogic, not maths. Learned, not guessed.
Neural networks learn through multiplication. LBNs learn through logic. The method is different. The discipline is the same. Training an LBN still requires data, a target, and deep learning. The difference is that ModelMill does most of it for you. Here’s what you need in order to get started with training logic-based AI.
Your data. Or everyone’s.
Good models begin with good data. That’s as true for LBNs as it is for any other class of AI.
ModelMill accepts structured data in CSV or JSON formats, and datasets up to 5 GB in size. If you don’t yet have a dataset to hand, you can train using open-source datasets including ModelMill’s inbuilt OpenML datasets. They enable you to begin training and benchmarking LBNs before your own data is even ready.
Compatible data input types: sensor, tabular, time series
ModelMill does the engineering.
AutoMLTraining an LBN from scratch should require a team of AI engineers, an extended development cycle, and significant compute. But it doesn’t. You upload your data, define where the LBN model needs to run, and ModelMill handles what a team of AI engineers would: auto-configuration, scaled training, deep-learning, candidate selection, and final deployment as a C code SDK ready to embed in your device.
ModelMill automates the process end to end. It’s been trained to do what Literal Labs’ engineers would do: configure, train, and benchmark hundreds of LBN candidates against your dataset and your target hardware. It then surfaces the best models for you to test and review.
You don’t need to build the model. You only need to know its purpose.
No new chips. No new capex.
Each LBN trained by ModelMill is delivered as a C code SDK. And inside that SDK sits the model itself, its inference engine, a build configuration, example code, and integration documentation.
That SDK is compatible with any 32-bit processor. ARM Cortex (32 and 64-bit), RISC-V, ESP, x86, and PowerPC are all supported. That means that LBNs run on the chips already inside your products, from sub-£1 microcontrollers to legacy industrial platforms that may have been in the field for a decade or more.
Neural networks typically demand new silicon. LBNs demand only that you have a processor. The one you already have will do.
Ready to train your first LBN?
Book a demo and bring your use case, your dataset, and your target hardware. We’ll show you what an LBN trained on your data looks like — and what it performs like — running on your device.