Week 1

Jan 01/02/2023

Holiday

Jan 01/03/2023

1. loe_coe_app
   1. document process of how to run this in Colab
   2. make a run and check (using Cox_PH and the utils package created)
2. migrate a few nbs to nbdev
   1. check nbs in the VS_internship folder
3. document the darts and neuralforecast types of variables (exogenous, future covariates, etc)
4. timeseries work
   • No reply from multiple forums where the question has been posted
   1. where are things currently
   2. next steps?
      1. wrt darts (join of inp_df and out_df dataframe and train model on that)
      2. wrt tsai (waiting on reply from tsai folks for the fix)
      3. wrt neuralforecast (error with custom dataloader) (unknown tuple error, only return X and try again)
5. doc for Melissa
   1. request review from Jake (email Jake, cc Sudhir)
6. reply Azure folks
7. move AI folder from Deven to Project, discuss with Sudhir
8. sphinx documentation
   1. loe_coe_app: ask Jake for some input
   2. chama22: even after running the running sphinx-apidoc for management/commands rst files are empty

Jan 01/04/2023

1. Reply to Sudhir's email
2. check the replies on the paper
3. document code formatting, isort and linting
   1. discuss use of pre-commit
   2. if code is not linted (or formatted), it will get unstaged
4. melissa's system up-to-date with python standards
5. deploy chama docs (check status → here)
6. Tried the join the dataframe and train the model (darts)
7. Need further reading on encoders and decoders
   1. Need to refresh understanding on timeseries forecasting models
   2. Most of the timeseries forecasting models are built on pytorch or pytorch_lightning
   3. List of papers/articles
      1. Temporal Fusion Transformer: A Primer on Deep Forecasting in Python
      2. Interpretable Deep Learning for Time Series Forecasting
      3. Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting
      4. Probabilistic Forecasting?
      5. Quantile loss in PyTorch?
      6. Get a refresher on creating dataloaders in PyTorch and PyTorchLightning
      7. Types of variables used in the forecasting: exogenous, past, future covariates

Next steps on timeseries front

1. Some quick reading from the Moroney book
   1. Moving average
   2. Vanilla (feed forward neural net)
   3. Use of keras-tuner to find ideal params
   4. LRSchedulers, etc
   5. Conv1D for filters, using (CNNs) for timeseries forecasting

Jan 01/05/2023

1. CHAMA
   1. update the IDEF0 (data flow) diagram (add the database interactions)
   2. summarize meeting notes
2. AI3 Hackathon
   1. revisit the work on the Elsevier API, need to scrape papers from containing various keywords related to AI and IoT, earthquakes, floods
3. Share update on time series front (results are in the copy of tsai notebook)

CHAMA meeting (9am - 11:30 am)

1. Went over meeting notes taken yesterday
2. Items have been created and assigned on the APTIM Chama project
3. Design decision for the config.json files
   1. We are not going to have a Django-based frontend up as the UI
   2. The database is only being used for documenting the runs at this point
4. Tasks in Deven's plate
   1. update the idef0 doc
   2. get Melissa's system caught up with the python dev standards and docs

References

1. https://www.mouser.com/blog/can-ai-systems-match-human-level-situational-awareness#:~:text=Situational%20Awareness%20in%20Artificial%20Intelligence&text=A%20combination%20of%20hybrid%20sensor,a%20human%20driver%20would%20perceive.
2. https://www.researchgate.net/publication/221298082_Natural_Language_Processing_to_the_Rescue_Extracting_Situational_Awareness_Tweets_During_Mass_Emergency
3. https://www.ckju.net/en/dossier/situational-awareness-what-it-and-why-it-matters-management-tool

AI3 Concept Paper Deadline (Jan 10 2023)

Things discussued with Sudhir

1. Ability to ingest data on the fly
2. Inclusion of metadata
3. Walky-talky used by first responders -- possible use of NLP in that case
4. Data categorization
   1. Labels like: human trapped, fire, etc
   2. Or existing labels
5. Earthquake scenarios
6. Semantic analysis
   1. Something similar was done in the RTEM
   2. Building data tags, etc
7. cascading effects
8. Knowledge extraction/ knowledge awareness
9. Event query language from Ramesh Jain
10. neo4j for knowledge graph
    1. write and execute queries to get knowledge out
11. data/knowledge extraction
12. data/knowledge organization
13. premise is that the proposed solution will have a pretrained knowledge graph/repo/database
    1. with the possibility of self-organizing map which can update itself on the fly
    2. update references

Jan 01/06/2023

1. Send the first draft out ASAP
2. Resources for anomaly detection
   1. https://github.com/hoya012/awesome-anomaly-detection
   2. https://unit8co.github.io/darts/generated_api/darts.ad.html
3. Made more updates and changes to the draft

Jan 01/08/2023

1. Had multiple discussions with Sudhir and one with Pavan
2. Discussing the minor nuances