### Interactive Model Prototyping in Visualization Space

Identifying and building models, and quantifying model parameters

-- Ove Daae Lampe and Helwig Hauser

### Who am I?

Ove Daae Lampe

Researcher at CMR Computing and PhD Student at UiB

# Interactive Model Prototyping in Visualization Space

### Model Building

is

• Understanding underlying behaviour of the data
• Abstracting data
• Quantifying controlling parameters

### Model Constructs

• Point Construct
• Linear Construct
• Exponential Construct

### Visualization Space

Model prototyping in Visualization Space analyzes visual representations of the data rather than the data.

In effect we rasterize the data once, and optimize the difference between this and a rasterized version of a model

### But WHY Visualization Space

• The data is rasterized once
• Optimize and fit multiple models efficiently
• Local models can be sketched
• Directly observe convergence / divergence and the Residual

### Interactive Convergence

Constant-step iterations while interacting. Newton's optimization applied afterwards.

Optimizing:

• Position / mean / μ
• Radius / std.dev / σ
• Scale / integral

### Workflow II

1. Visualize and Observe
2. Sketch and Fit
3. Externalize and Subtract
4. Iterate

### Test Dataset

• A(x) = N(0,1)
• C(x) = 0.05N(1,0.2)
• B(x) = A(x)+C(x)

### Step 3. Externalize ...

Modeled:
• μ = (0.02,0.01)
• σ = (1.035,1.04)
Reference:
• μ = (0.0,0.0)
• σ = (1.0,1.0)

### Step 4. Iterate

Modeled:
• μ = (1.004,1.004)
• σ = (0.1989,0.2015)
Reference:
• μ = (1.0,1.0)
• σ = (0.2,0.2)

# Domain

## Example

### Drilling Data II

Torque measured in kN.m over time in seconds.

### Externalized Result

Result: a quantitative plot of mean torque over depth and error (one standard deviation ≅ 68% of the measurements).

### Summary

• A workflow for sketching models in visualization space
• Combining sketching and automated fitting
• Externalization through quantification
• Efficient algorithm