White Papers

8 Deep Learning using iAbra stack on Dell EMC PowerEdge Servers with Intel technology
2.2 iAbra AI Use Case Qualification
Embedded Use Case Characteristics
○ Low Size, Weight and Power
○ Type Approval or
○ Plan of Record Compliant
○ Low Failure Rates
○ Environmental Survival
○ security
Network Creation Use Case Characteristics
○ Embedded End Use Case
○ Abstracted Training Platform
○ Non-Data Scientist Users
○ Reduced time to Solution
○ Smaller more targeted networks (e.g. sub 1000 neurons)
○ Training to Inference Fidelity
2.2.1 How are the use cases addressed?
iAbra’s tools provide Machine Learning (ML) optimized for FPGA Inference. While the resurgence in ML has
been facilitated by the growth in computing power attributable to Moore’s Law and specialized GPU hardware
and associated software, the use of the subsequent models for processing data is another matter.
Artificial intelligence inference, or the processing of data with a machine learnt model might not be possible
with real work constraints. In the military defense realm these can be numerous and include silicon size,
weight and power (SWaP) along with security, environmental survival and integration with plan of record.
iAbra’s focus on tools that create AI models for inference on FPGA seeks to ensure that real world constraints
do not prevent the use of AI. Our perspective is that AI use cases are multiple not just outside of the data
center, but at the edge, where bandwidth might be limited, or network access denied. When data connectivity
constraints are coupled with low SWaP requirements, FPGAs are a rational choice.
Neural Network creation and training is often equal parts art and Science. iAbra’s experience of the
fragmented ecosystems relating to ML and AI while developing FPGA based models for inference, motivated
the creation of our end to end tool chain which automates the network creation and training. This enables a
user to create a neural network, compiled to an FPGA for AI inference, using an abstracted interface. This
enables an analyst, without specific ML skills, to create a model that can not only amplify and augment the
process of extracting utility from data in support of defined outcomes but be used on FPGA silicon for down
range implementation not practicable with some other silicon.
2.2.2 Implementation
• iAbra’s tool flow takes raw training data and creates a compact application specific NN