Data Science Initiatives at Poli - Universidade de São Paulo · Large Scale Acoustic Sensing...

Preview:

Citation preview

Data Science Initiatives at Poli

Fabio Cozman

Escola Politécnica (Poli)

Poli

• Started in 1893; now offers 15 courses.

• About 5000 undergraduate students; about 1500 graduate students.

A Very Diverse School

• Computer, Electrical, Mechanical, Naval, Production, Civil, Chemical, Materials, Mining, Petroleum Engineering.

• Data Science efforts in all corners.

Large Scale Acoustic Sensing

• Mapping animals (for instance owls) at Mata Atlantica.

– Data: acoustic traces of animals in wilderness.

– Related application: monitoring forest recovery.

• Lab of Acoustic and Environment – LACMAM (Linilson Padovese)

Air Quality Monitoring

• Getting data from all stations in São Paulo.

• Clustering areas, predicting pollution levels.

• Joint effort USP-CETESB-FAPESP (Roberto Guardani)

Civil Construction Control

• Monitoring team and production through RFID systems.

• Goal: to predict unexpected changes in construction (modeled by Petri nets with stochastic elements).

• PCC-Poli (Fabiano Correa)

Digital Twins in Oil Exploration • Long term goal: digital Twin fed with data from oil

platform (with autonomy…). • Current effort:

archor failure identification, platform movement prediction, natural language for report search.

• Petrobrás-Poli (Edson Gomi, Eduardo Tannuri, Alexandre Simos, Anna Reali)

To do research and to train researchers, improving the data science ecosystem in the country.

• Initiatives: – Scholarship program.

– Seminars, workshops.

– Hackatons, competition teams.

– IA Cidadã.

Research topics

• Next generation conversational systems.

• Transfer learning.

• Market prediction.

• Recommendation systems.

• Explainable AI.

Interpretability: A Necessity

• Business necessity, and also legal necessity.

… but depends on user, and is subjective; depends on task...

Model-agnostic Strategies

Example: Knowledge-base Completion

• Take knowledge-base NELL186; best completion methods by embeddings.

• How to justify a completion?

Concluding…

• Data Science is big at Poli – a very diverse place.

• Much data, many applications.

• Many recent initiatives given surge of practical and theoretical interest.