Title: Bayesian networks and how they can
1Bayesian networks and how they can help us to
explore fish species interaction in the Northern
gulf of St Lawrence
- Dr Allan Tucker
- Centre for Intelligent Data Analysis
- Brunel University
- West London
- UK
2Talk Outline
-
- Introduce myself and research group
- Introduce Machine Learning
- Describe Bayesian network models
- Document some preliminary results on fish
population data - Conclusions
3Who Am I?
- Research Lecturer at Brunel University, West
London - Member of Centre for IDA (est 1994)
X
4What is the ?
- Over 25 members (academics, postdocs, and PhDs)
with diverse backgrounds (e.g. maths, statistics,
computing, biology, engineering) - Over 140 journal publications a dozen research
council grants since 2001 - Many collaborating partners in UK, Europe, China
and USA - Bi Annual Symposia in Europe
5Some Previous Work in
- Machine Learning and Temporal Analysis
- Oil Refinery Models
- Forecasting
- Explanation
- Medical Data Retinal (Visual Field)
- Screening
- Forecasting
- Bioinformatics
- Gene Clusters
- Gene Regulatory Networks
6Some Previous Work in
7Part 1
What is Machine Learning?
8What is Machine Learning?
- (and why not statistics?)
- Data oriented
- Extracting useful info from data
- As automated as possible
- Useful when lots of data and little theory
- Making predictions about the future
9What Can we do with ML?
-
- Classification and Clustering
- Feature Selection
- Prediction and Forecasting
- Identifying Structure in Data
10E.g. Classification
-
- Given some labelled data (supervised)
- Build a model to allow us to classify other
unlabelled data - e.g. A doctor diagnosing a patient based upon
previous cases
11Classification e.g. medical
- Scatterplot of patients
- 2 variables
- Measurement of expression of 2 genes
12Classification
- How do we classify them?
- Nearest Neighbour / Linear / Complex Fn?
13Classification
- Trivial case with Cod and Shrimp Data
14The Data
- Northern Gulf (region a)
- Two ships (Needler and Hammond) combined by
normalising according to overlap year - Multivariate Spatial Time Series (short)
- Missing Data
15Background
- Northern Gulf considered to be one ecosystem /
fish community - Quite heavily fished until about 1990
- Most fish populations collapsed since
- Some say that moved to an alternative stable
state and unlikely to come back to cod dominated
community without some chance event beyond human
control. - Lots of speculation
- cold water
- large increases in population of predators.
- Examine nature and strength of interactions
between species in the two periods. - Ask what if ? questions
- For other parts of community to recover, we would
need cod to have X strength of interaction with Y
number of other species?
16ML for Northern Gulf Data
- Network building
- knowledge and data of interactions
- Feature Selection for Classification of relevant
species to the cod collapse - State Space / Dynamic models for predicting
populations - Hidden variable analysis
17Part 2
Bayesian Networks for Machine Learning
18Bayesian Networks
- Method to model a domain using probabilities
- Easily interpreted by non-statisticians
- Can be used to combine existing knowledge with
data - Essentially use independence assumptions to
model the joint distribution of a domain
19Bayesian Networks
- Simple 2 variable Joint Distribution
- can use it to ask many useful questions
- but requires kN probabilities
P(Collapse1, Collapse2)
Species2 Species2
Species1 0.89 0.01
Species1 0.03 0.07
20Bayesian Network for Toy Domain
SpeciesA
SpeciesB
P(A)
P(B)
.001
.002
A B P(C)
T T .95
T F .94
SpeciesC
F T .29
F F .001
C P(E)
C P(D)
T .70
T .90
F .01
F .05
SpeciesD
SpeciesE
21Bayesian Networks
- Bayesian Network Demo
- Species_Net
- Use algorithms to learn structure and parameters
from data - Or build by hand (priors)
- Also continuous nodes (density functions)
22Informative Priors
- To build BNs we can also use prior structures
and probabilities - These are then updated with data
- Usually uniform (equal probability)
- Informative Priors used to incorporate existing
knowledge into BNs
23Bayesian Networks for Classification Feature
Selection
- Node that represents the class label attached to
the data
24Dynamic Bayesian Networks for Forecasting
- Nodes represent variables at distinct time
slices - Links between nodes over time
- Can be used to forecast into the future
- Species_Dynamic_Net
25Hidden Markov Models
- Like a DBN but with hidden nodes
- Often used to model sequences
HT-1
HT
OT-1
OT
26Typical Algorithms for HMMs
- Given an observed sequence and a model, how do
we compute its probability given the model? - Given the observed sequence and the model, how
do we choose an optimal hidden state sequence? - How do we adjust the model parameters to
maximise the probability of the observed sequence
given the model?
27Summary
- Different learning tasks can be used to solve
real world problems - Machine Learning techniques useful when lots of
data and lots of gaps in knowledge - Bayesian Networks probabilistic framework that
can perform most key ML tasks - Also transparent can incorporate expert
knowledge
28Part 3
Some Preliminary Results on Northern Gulf Data
29Expert Knowledge
- Ask marine biologists to generate matrices of
expected relationships - Can be used to compare models learnt from data
- Also to be used as priors to improve model
quality
30Results Expert networks
31Results Data networks (BN from correlation)
- 85 conf. imputed from 70 data
- Warning data quality, spurious relations
(Eel pout / Ocean Sun Fish)
Witch Flounder
(Lumpfish)
Cod
Haddock
Shrimp
(Silver Hake)
(Atlantic soft pout / Bristlemouths)
32Example DBN
- Lets look at an example DBN
- NGulfDynamic - range
- Structure Encoded by knowledge
- Updated by data
- Explore with queries
- Supported by previous knowledge
- In the Northern gulf of st. Lawrence, cod (code
438) and redfish (792,793,794,795,796) collapsed
to very low levels in the mid 1990s. Subsequently
the shrimp (8111) increased greatly in biomass so
one will see this signal in the data. It is
hypothesised that these are exclusive community
states where you never get high abundance of both
at the same time owing to predatory interactions.
33Feature Selection
- Given that we know that from 1990 the cod
population collapsed - Can we apply Feature Selection to see what
species characterise this collapse - Learn BN and apply CV
34Results 7 Feature Selection with Bootstrap
Filter method using Log Likelihood
Wrapper method using BNs
Redfish
35Results Feature Selection
- Change in Correlation of interactions between
cod and high ranking species before and after
1990
36Dynamic Models
- Given that the data is a time-series
- Can we build dynamic models to forecast future
states? - Can we use HMM to classify the time-series?
37Multivariate Time Series
- N Gulf is process measured over time
- Autoregressive Correlation Function
- (here cod)
- Cross Correlation Function
- (here hake to cod)
ACF
CCF
38Results 3 Fitting Dynamic Models
- HMM Expert with CCF gt 0.3 (maxlag 5)
LSS 8.3237
39Results 3 Fitting Dynamic Models
- Learning DBN from CCF data
LSS 5.0106
Fluctuation Early Indicator of Collapse?
40Results 4 Examining DBN Net
Hakes
Redfish
Cod
Haddock
Witch Flounder
White Hake
Thorny Skate
Shrimp
41Results 5 Fitting Dynamic Models
- Learning DBN from Expert biased CCF data CCF gt
0.5 (maxlag5)
LSS 6.1326
42Results 6 Examining DBN Net
- Data Biased Expert Dynamic Links
Cod
Herring
Witch Flounder
Mackerel / Capelin
43Results 7 Linear Dynamic System
- Instead of hidden state, continuous var
- Could be interpreted as measure of fishing?
Predator population (e.g. seals)? Water
temperature?
1987
1991
1997
1984
44Conclusions
- Hopefully conveyed the broad idea of machine
learning - Shown how it can be used to help analyse data
like fish population data - Potentially applicable to other data studied
here at MLI
45Potential Projects
- Spatio-Temporal Analysis
- Use Spatio-Temporal BNs to model fish stock data.
Nodes would represent species in specific
regions - Combining Expert Knowledge and Data for improved
Prediction - Looking for Un/Stable States and the factors that
influence them - Machine Learning Techniques for other Data
generated here at MLI
46E.G. Spatial Analysis
- Spatial Bayesian Network Analysis
- NGulfCodSpatial
47AcknowledgementsDaniel Duplisea for inviting
me
Any Questions?