21st International Joint Conference on Artificial Intelligence (IJCAI), Pasadena

Machine Learning in Ecosystem Informatics and Sustainability

introducer: Carlos Guestrin, Computer Science Department, Carnegie Mellon University
author: Thomas Dietterich, School of Electrical Engineering and Computer Science, Oregon State University
published: July 22, 2009, recorded: July 2009, views: 390

Slides

Slides
0:00	- Introduction
2:47	Machine Learning in Ecosystem Informatics and Sustainability
3:16	Threats to the Biosphere
4:05	Needed: Robust Optimal Policy Based on Sound Science
6:03	A Limiting Factor: Ecological Data
8:01	Ecosystem Sciences
11:21	Data Pipeline
11:44	Data Pipeline - Optimal Sensor Placement
11:46	Optimal Sensor Placement for Environmental Data Collection
12:54	Data Pipeline - Detectability Errors / Noise Sampling Bias
13:35	Sampling Bias: ebird.org
15:04	Detectability
16:48	Data Pipeline - Species classification
17:18	The BugID Project: Rapid Throughput Arthropod Counting
19:07	Data Pipeline - Sensor failures, Networking failures
19:43	Multi-Sensor Anomaly Detection
20:21	Data Pipeline - Species distribution models
20:50	Species Distribution Models
22:44	Plants in Victoria
23:30	Data Pipeline - Optimal Sensor Placement
23:57	Robust Reserve Design
27:51	Outline
28:01	Automated Rapid-Throughput Arthropod Population Counting
28:33	Application 1: Stonefly populations in freshwater streams
29:51	Application 2: Small arthropods in soil: “soil mesofauna”
30:18	Application 3: Freshwater Zooplankton
30:24	Image Capture Apparatus
31:15	Robotic Extraction of Specimens
31:55	Computer Vision Challenges (1)
32:20	Computer Vision Challenges (2)
32:46	Computer Vision Challenges (3)
34:22	Machine Learning
34:39	Region-Based Approaches:Convert Image to Bag of Patches
36:52	Defining the Patches: Interest Region Detectors
37:57	Representing the Patches:SIFT (Lowe, 1999)
39:47	Classify Bag of Patches Method 1: Visual Dictionaries
40:46	Learn visual dictionary via clustering
41:31	Classify Bag of Patches Method 1: Visual Dictionaries
41:42	Classify Bag of Patches Method 2: Multiple-Instance Classifier
43:37	Improved Multiple-Instance Classification
43:54	Classify Bag of Patches Method 2: Multiple-Instance Classifier
43:58	Improved Multiple-Instance Classification
45:19	Classify Bag of Patches Voted Evidence Trees
45:59	Theorem: Voting Evidence is Better than Voting Decisions
47:04	Ensemble Learning
47:16	Final Classifier: Stacked Random Forests
48:20	Experimental Study 9 Taxa of Stoneflies
49:05	STONEFLY9 Dataset
49:24	Comparison of Methods
49:47	Issues with Visual Dictionaries
51:27	Next Steps
52:04	Outline
52:12	Upper Lookout Met. Stationthermometers at 1.5, (1)
52:49	Upper Lookout Met. Stationthermometers at 1.5, (2)
55:27	Approach: Learn a Very Accurate Model of Normal Behavior
56:02	Single Sensor Bayesian Network Model (1)
56:54	Single Sensor Bayesian Network Model (2)
57:37	Assessment
58:20	Multiple Sensors
58:31	Example: SensorScope (EPFL, Switzerland)
58:59	Multi-Sensor Anomaly Detection
59:32	Multiple Sensor Evaluation
60:28	Institute for Computational Sustainability
62:06	Summary
62:36	For More Information…
63:50	Acknowledgements
64:00	Questions?

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.

Lecture popularity: You need to login to cast your vote.

Description

Ecosystem Informatics brings together mathematical and computational tools to address scientific and policy challenges in the ecosystem sciences. These challenges include novel sensors for collecting data, algorithms for automated data cleaning, learning methods for building statistical models from data and for fitting mechanistic models to data, and algorithms for designing optimal policies for biosphere management. This talk will describe recent work on the first two of these---new devices for automated arthropod population counting and linear Gaussian DBNs for automated cleaning of sensor network data. It will also give examples of open problems along the whole spectrum from sensors to policies.