Welcome to the GEOMAR AI Seminar Series

Everardo González (egonzalez@geomar.de)

Naveen Parameswaran (nparameswaran@geomar.de)

\[N=R_{*}\cdot f_{\mathrm{p}}\cdot n_{\mathrm{e}}\cdot f_{\mathrm{l}}\cdot f_{\mathrm{i}}\cdot f_{\mathrm{c}}\cdot L\]

"But where is everybody?"

— Enrico Fermi, 1950

Chpt. 1

Follow the Science

Uncertainty Quantification and Information Gain

One big question:

"Where to Sample Next?"

"Information Gain"

(Kullback–Leibler Divergence)

$D_{KL}(P\|Q) =$


$-\sum\limits_{x\in X}$

$p(x)$

$\log($

$q(x)$

$) + \sum\limits_{x\in X}$

$p(x)$

$\log($

$p(x)$

$)$


"Information Gain"

(Kullback–Leibler Divergence)

$D_{KL}(P\|Q) =$


$-\sum\limits_{x\in X}$

$p(x)$

$\log($

$q(x)$

$) + \sum\limits_{x\in X}$

$p(x)$

$\log($

$p(x)$

$)$


Observation Probability

Prediction Probability

Features

~400 Global Maps

Labels

~5000 Total Organic Carbon Measurements

Neural Network with DropOut

Monte Carlo Drop Out

Chpt. 2

Follow the Techniques

Shap & Uncertainty Quantification

One day...

Previous Works: Uncertainty Quantification, SHAP, ...

InfoSHAP: Uncertainty Quantification x SHAP

🤔

Chpt. 3

Follow the Data

Acoustic Embeddings for Underwater Ecosystem Monitoring

The Data

~50h recordings in German North Sea coastal waters

The Data

5 sec. snippets ⇒ ~40,000 labeled (!!!) datapoints

~20 unique feature combinations ⇒ classes

Solution: Word Embeddings!

Can we do that for underwater soundscapes?

The Model

contrastive learning with MobileNet 2

Loading visualisation...

Audio Sample Info

Legend

Chpt. 4

Follow your Colleagues

AI x Ocean Science: GEOMAR ML Seminar Series

AI x Ocean Science:
GEOMAR AI Seminar Series

  • Meet other people working on ML
  • Open discussion, transdisciplinary ideas exchange
  • Get inspired, do great science

Fin

Questions? Comments?

egonzalez@geomar.de