search

mlcommons / training_policies

Issues related to MLPerf™ training policies, including rules and suggested changes
https://mlcommons.org/en/groups/training
Apache License 2.0
91 stars 65 forks source link

3, 1 Paragraph Summaries #402

Open bitfort opened 3 years ago

bitfort commented 3 years ago

Relevant Applications, key technical challenge, etc.1-paragraph for each model describing why we care. (Add to reference contributing and vision):

bitfort commented 3 years ago

AI(NV) - send out example of the paragraph for reference owners to use a starting point.

TheKanter commented 3 years ago

Example 1 from mobile:

Image classification picks the best label to describe an input image and is commonly used for photo search and text extraction. The MobileNetEdgeTPU reference model is evaluated on the ImageNet 2012 validation dataset and requires ?? Top-1 accuracy (app uses a different dataset).

The MobileNetEdgeTPU network is a variant of the MobileNet-v2 family that is optimized for low-latency and mobile accelerators. The MobileNetEdgeTPU model architecture is based on convolutional layers with inverted residuals and linear bottlenecks, similar to MobileNet v2, but is optimized by introducing fused inverted bottleneck convolutions to improve hardware utilization, and removing hard-swish and sqeeze-and-excite blocks.

bitfort commented 3 years ago

Examples:

Single Shot MultiBox Detector (SSD) is an object detection network. For an input image, the network outputs a set of bounding boxes around the detected objects, along with their classes. For example:

https://upload.wikimedia.org/wikipedia/commons/3/38/Detected-with-YOLO--Schreibtisch-mit-Objekten.jpg

SSD is a one-stage detector, both localization and classification are done in a single pass of the network. This allows for a faster inference than region proposal network (RPN) based networks, making it more suited for real time applications like automotive and low power devices like mobile phones.

bitfort commented 3 years ago

SWG:

We want to explain the model and benchmark 3 times. We are now looking for 3 paragraphs for 3 audiences: layperson, non-ML technical person (CS undergrad / PM / former engineer), ML-technical (e.g. engineer/researcher).