ml-update-demo: Update from a faulty model to a working one

Add a modified tflite file that produces no inference results
at runtime. This file can be used to build an application that
is fully functional, but the ML inference is unsuccessful in
detecting any keyword. Applying an OTA ML model update in this
state is very visible since the updated model does work as
expected (detects keywords from audio samples).

Signed-off-by: Filip Jagodzinski <[email protected]>
Signed-off-by: Ahmed Ismail <[email protected]>

Author: Filip Jagodzinski

applications/keyword_detection/ml-model-update-demo/faulty_kws_micronet_m.tflite

@@ -108,4 +108,96 @@ only differences are:
 
 Now, start the Keyword-Detection example, and observe the ML model update.
 
+#### Making the ML model update demo more appealing
+
+Although it is enough to observe the ML model component version bump, a more
+demonstrative output can be obtained with minimal effort. Follow the (optional)
+steps below to run the Keyword-Detection example with a modified faulty model, unable
+to detect any keyword, deploy the OTA update of the ML model, after which correct
+ML inference results can be observed after the update is complete.
+
+1. Save the update image with a correctly working ML model.
+
+    By default, the keyword-detection example is built with a fully-functional
+    ML model, fetched from the [ML-zoo][kws-model]. If you have already built
+    the application, the signed model update is available in the build
+    directory. Back it up together with its signature string.
+
+    ```bash
+    cp build/keyword-detection-model-update_signed.bin build/model-update-signature.txt applications/keyword_detection/ml-model-update-demo
+    ```
+
+1. Alter the ML model artifacts in the build dir.
+
+    A modified model is available in
+    `../../../applications/keyword_detection/ml-model-update-demo/faulty_kws_micronet_m.tflite`. Compile it with
+    Vela and replace the original tflite file.
+
+    ```bash
+    source build/mlek_resources_downloaded/env/bin/activate && vela applications/keyword_detection/ml-model-update-demo/faulty_kws_micronet_m.tflite --accelerator-config=ethos-u55-128 --optimise Performance --config components/ai/ml_embedded_evaluation_kit/library/scripts/vela/default_vela.ini --memory-mode=Shared_Sram --system-config=Ethos_U55_High_End_Embedded --output-dir=applications/keyword_detection/ml-model-update-demo --arena-cache-size=2097152
+    cp applications/keyword_detection/ml-model-update-demo/faulty_kws_micronet_m_vela.tflite build/mlek_resources_downloaded/kws/kws_micronet_m_vela_H128.tflite
+    ```
+
+1. Build the application with the modified model.
+
+    Simply run the build command mentioned in the [Build
+    instructions](#build-instructions) section.
+
+1. Run the Keyword-Detection application and confirm that no keywords are
+detected. Then stop the application.
+
+1. Deploy an AWS OTA job with the functional ML model from the
+`../../../applications/keyword_detection/ml-model-update-demo` dir.
+
+1. Start the Keyword-Detection example again, let it update the ML model, and
+detect keywords correctly again.
+
+
+Before the ML Model update:
+
+```
+(...)
+58 10031 [ML_TASK] [INFO] Running inference on an audio clip in local memory
+59 10058 [OTA Agent Task] [INFO] Current State=[WaitingForJob], Event=[ReceivedJobDocument], New state=[CreatingFile]
+60 10103 [ML_TASK] [INFO] ML UNKNOWN
+61 10109 [ML_MQTT] [INFO] Attempting to publish (_unknown_) to the MQTT topic MyThing_eu_central_1/ml/inference.
+62 10128 [ML_TASK] [INFO] For timestamp: 0.000000 (inference #: 0); label: <none>; threshold: 0.000000
+63 10168 [ML_TASK] [INFO] For timestamp: 0.500000 (inference #: 1); label: <none>; threshold: 0.000000
+64 10208 [ML_TASK] [INFO] For timestamp: 1.000000 (inference #: 2); label: <none>; threshold: 0.000000
+65 10248 [ML_TASK] [INFO] For timestamp: 1.500000 (inference #: 3); label: <none>; threshold: 0.000000
+66 10288 [ML_TASK] [INFO] For timestamp: 2.000000 (inference #: 4); label: <none>; threshold: 0.000000
+67 10328 [ML_TASK] [INFO] For timestamp: 2.500000 (inference #: 5); label: <none>; threshold: 0.000000
+68 10368 [ML_TASK] [INFO] For timestamp: 3.000000 (inference #: 6); label: <none>; threshold: 0.000000
+69 10408 [ML_TASK] [INFO] For timestamp: 3.500000 (inference #: 7); label: <none>; threshold: 0.000000
+```
+
+After the ML Model update:
+
+```
+(...)
+57 10000 [OTA Agent Task] [INFO] In self test mode.
+58 10009 [OTA Agent Task] [INFO] New image has a higher version number than the current image: New image version=0.0.42, Previous image version=0.0.11
+59 10034 [OTA Agent Task] [INFO] Image version is valid: Begin testing file: File ID=0
+(...)
+75 12259 [OTA Agent Task] [INFO] New image validation succeeded in self test mode.
+(...)
+95 14005 [ML_TASK] [INFO] Running inference on an audio clip in local memory
+96 14032 [OTA Agent Task] [INFO] Current State=[WaitingForJob], Event=[ReceivedJobDocument], New state=[CreatingFile]
+97 14078 [ML_TASK] [INFO] ML_HEARD_ON
+98 14084 [ML_MQTT] [INFO] Attempting to publish (on) to the MQTT topic MyThing_eu_central_1/ml/inference.
+99 14102 [ML_TASK] [INFO] For timestamp: 0.000000 (inference #: 0); label: on, score: 0.996127; threshold: 0.700000
+100 14144 [ML_TASK] [INFO] For timestamp: 0.500000 (inference #: 1); label: on, score: 0.962542; threshold: 0.700000
+101 14186 [ML_TASK] [INFO] ML UNKNOWN
+102 14192 [ML_TASK] [INFO] For timestamp: 1.000000 (inference #: 2); label: <none>; threshold: 0.000000
+103 14232 [ML_TASK] [INFO] ML_HEARD_OFF
+104 14239 [ML_TASK] [INFO] For timestamp: 1.500000 (inference #: 3); label: off, score: 0.999030; threshold: 0.700000
+105 14281 [ML_TASK] [INFO] ML UNKNOWN
+106 14287 [ML_TASK] [INFO] For timestamp: 2.000000 (inference #: 4); label: <none>; threshold: 0.000000
+107 14328 [ML_TASK] [INFO] For timestamp: 2.500000 (inference #: 5); label: <none>; threshold: 0.000000
+108 14368 [ML_TASK] [INFO] ML_HEARD_GO
+109 14375 [ML_TASK] [INFO] For timestamp: 3.000000 (inference #: 6); label: go, score: 0.998854; threshold: 0.700000
+110 14417 [ML_TASK] [INFO] ML UNKNOWN
+111 14423 [ML_TASK] [INFO] For timestamp: 3.500000 (inference #: 7); label: <none>; threshold: 0.000000
+```
+
 [kws-model]: https://github.com/ARM-software/ML-zoo/tree/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8

applications/keyword_detection/ml-model-update-demo/kws_micronet_m.tflite

@@ -92,7 +92,7 @@
               "properties": {
                 "path": {
                   "type": "string",
-                  "pattern": "(gitmodules|CODEOWNERS|doxyfile_options|gitlint|license\\.schema|^(?!CMakeLists).+\\.txt|.gitkeep|.astylerc)$"
+                  "pattern": "(gitmodules|CODEOWNERS|doxyfile_options|gitlint|license\\.schema|^(?!CMakeLists).+\\.txt|.gitkeep|.astylerc|.tflite)$"
                 },
                 "type": {
                   "type": "string",