## conv2d1 (indirect previous layer)

Weights conencting this unit to the 50 units it's most strongly connected to (by L2 norm). Units organized by factorizing the weight matrix if weight factorization explains significant variance.

## conv2d2 (indirect previous layer)

Weights conencting this unit to the 50 units it's most strongly connected to (by L2 norm). Units organized by factorizing the weight matrix if weight factorization explains significant variance.

## mixed3a (previous layer)

Weights conencting this unit to the 50 units it's most strongly connected to (by L2 norm). Units organized by factorizing the weight matrix if weight factorization explains significant variance.

## mixed4a (next layer)

Weights conencting this unit to the 50 units it's most strongly connected to (by L2 norm). Units organized by factorizing the weight matrix if weight factorization explains significant variance.

## mixed4b (indirect successor layer)

Weights conencting this unit to the 50 units it's most strongly connected to (by L2 norm). Units organized by factorizing the weight matrix if weight factorization explains significant variance.

## mixed4c (indirect successor layer)

Weights conencting this unit to the 50 units it's most strongly connected to (by L2 norm). Units organized by factorizing the weight matrix if weight factorization explains significant variance.