Week 2

May 05/11/23

1. AI3 hackathon
   1. Given a stream of values, create a simple classifier like SVM, XGBoost or LightGBM
   2. Sample prediction script which Jake can directly use for Prediction
      1. Model size estimates:
         1. 6.7B param models renders the laptop un-usable
         2. The largest model that we can use on a CPU run-time is the 1.3B model
         3. For the Raspberry Pi, I would suggest we use the smallest (125M param) OPT model
         4. The model with 350M params works as well.
   3. Download the new SD1 videos
   4. JSON Standardization
   5. Abbreviations API

May 05/10/23

Vacation day

May 05/09/23

1. New jobs to submit for SD1 with Set C (71 Sanitary and combined sewers videos from Set A and Set B)
   1. A50 Y:0.8 C ML DF FB
   2. A50 Y:0.8 C ML DF VB
   3. A50 Y:0.8 C ML FB
   4. A50 Y:0.8 C ML VB
   5. A50 [X:1.0 (A+B) + Z:1.0 (A)] ML
   6. Output of v. → finetune Y:0.8 C ML FB
   7. Output of v. → finetune Y:0.8 C ML VB
   8. A50 [X:1.0 (A+B) + Z:1.0 (A) + Y:0.8 C] SL FB
   9. A50 [X:1.0 (A+B) + Z:1.0 (A) + Y:0.8 C] SL VB
   10. Output of viii. (stage 1) + output of vi. (stage 2)
   11. Output of ix. (stage 1) + output of vii. (stage 2)

May 05/08/23