We’re thrilled to introduce a major update to the Edge Impulse platform: Users can now seamlessly ingest tabular data, directly into their Edge Impulse projects — along with full support for BYOM (Bring Your Own Model). This enhancement empowers users who have pre-processed their data outside of the platform, or who have non-time-series numerical data, offering more flexibility in integrating data for model development.
Previous visualization for non-time-series data (left) vs. new visualization with tabular data release (right)
Why This Matters
Edge Impulse has been a powerful platform for processing raw data like time-series and images, and now we’re taking it even further. With this update, we’re empowering users by enabling seamless integration of pre-processed or non time series data, giving them more flexibility in how they work. Whether you process data externally or face restrictions with raw data, this update makes it even easier to leverage Edge Impulse for all your data handling and model training needs.
This is especially beneficial in the following scenarios:
- Third-party IP restrictions: Some devices or services only provide processed features, restricting access to raw data. This update allows users to still utilize Edge Impulse even when raw data isn't available.
- Enterprise data pipelines: Organizations can process data as part of their internal pipelines and upload the processed features directly to Edge Impulse, streamlining the process.
- Bandwidth limitations: Large raw datasets can be difficult to manage in low-bandwidth environments. Ingesting pre-processed features reduces the burden of data transmission.
- External feature extraction: Users who have already defined feature extraction methods outside of our platform can now seamlessly ingest those features for model development without having to repeat the process.
- Complex tabular data: This update introduces native support for tabular data, simplifying how arbitrary, non-time-series samples can be processed and used for training.
- Sensitive data: In industries like healthcare, where privacy is paramount, users can upload processed or anonymized features to avoid handling raw sensitive data while still leveraging the platform's capabilities.
- Regulatory compliance: In industries with strict legal frameworks like GDPR or HIPAA, companies can remain compliant by sharing only processed features, rather than raw data.
A New Era for Tabular Data
Along with pre-processed feature ingestion, Edge Impulse now natively supports tabular data processing, making it easier to work with data that doesn’t rely on a time axis. This brings a more intuitive way to view, manipulate, and train with non-time-series datasets. For example:
- Spectroscopy: Where data samples consist of energy measurements versus frequency.
- X-ray diffraction: Where intensity is measured against diffraction angles.
- Meteorology: Where conditions like temperature, wind speed, and pressure are classified based on their relationships, rather than temporal trends.
With this update, users can train models on tabular data directly within the platform, or leverage tools like our Flatten block to reduce the dimensionality of complex datasets.
Also, check out an example project for HRV analysis!
Conclusion
By supporting pre-processed feature and native tabular data ingestion, Edge Impulse expands its platform to meet a wider variety of use cases. Whether you're dealing with IP limitations, large data pipelines, or sensitive data, this update provides the flexibility to bring data into Edge Impulse in ways that suit your needs. Train directly on pre-processed features, optimize workflows, and develop more sophisticated models with ease.
We’re excited to see what you build with this new powerful data acquisition tool in Edge Impulse, tag us on social media @EdgeImpulse or tell us more on our forum!