On Darwinism and Side Channel Resistance RHUL 2013 Talk David Naccache.
-
Upload
earl-moody -
Category
Documents
-
view
216 -
download
0
Transcript of On Darwinism and Side Channel Resistance RHUL 2013 Talk David Naccache.
Müllerian Mimicry
“A natural phenomenon in which two or more poisonous species, that may or may not be related and share one or more common predators, have come to mimic each other's warning signals.”
(Named after German naturalist Fritz Müller, who proposed the concept in 1878).
Digital Müllerian Similarity
Rules: attack and defense have the same rights:• Both Have the same knowledge about the system• Both can experiment as much as they wish with the system.• Instead of transmitting bits, the defender sacrifices part of
his bandwidth to buy similarity.
𝑘 camouflage value
The -bit number
The Game
𝑓 𝑑 (𝑡 )𝑘 ,𝑣
Defender’s goal: for each select a .
For the values must be such that: is as similar as possible to
The Game
𝑓 𝑑 (𝑡 )𝑘 ,𝑣
Defender’s goal: for each select a .
For the values must be such that: is as similar as possible to
Be as discreet as possible and let noise do the rest of the cover-up work!
A Small ExampleConsider bits.We devote 2 bits to payload data (i.e. ) and the 2 other bits to camouflage (i.e. )
A Small ExampleConsider bits.We devote 2 bits to payload data (i.e. 0,1,2,3) and the 2 other bits to camouflage (i.e. )
We call these « colors »
A Small ExampleWe have sixteen values belonging to four color groups.We need to select one representative in each group
10001001101010111100110111101111
0000 0001001000110100010101100111
A Small ExampleWe have sixteen values belonging to four color groups.We need to select one representative in each group
10001001101010111100110111101111
0000 0001001000110100010101100111
speciesmelopomene
speciescydna
specieserato
speciessapho
Apply Natural SelectionWe have sixteen values belonging to four color groups.We need to select one representative in each group
10001001101010111100110111101111
0000 0001001000110100010101100111
speciesmelopomene
speciescydna
specieserato
speciessapho
Apply Natural Selection
10001001101010111100110111101111
0000 0001001000110100010101100111
speciesmelopomene
speciescydna
specieserato
speciessapho
Criterion: Analogy: MSB = basic genotype, LSB = mutations
How to Find the Good One?Simulating Mutations is easy, measuring similarity is not! Assume that a simplistic blue jay has only one criterion: the global intensity of yellow in the butterfly.
1000 31001 91010 71011 41100 31101 91110 41111 2
0000 0 0001 00010 10011 30100 00101 10110 10111 4
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Yes
Current optimum
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Yes
Current optimum
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
No
Current optimum
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Yes
Current optimum
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Yes
Current optimum
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Current optimum
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Current optimum
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Current optimum
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Current optimum
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Current optimum
No
Sort Butterflies by Yellowness
0 1 2 3 4 5 6 7 8 9
Problem: Find the shortest segment containing all colorsA quadratic-time algorithm: just scan and keep an optimum.
Current optimum
No
If The Blue Jay is More Perceptive?Cyanocitta cristata oculus-laseris perceives both size and yellow levels.
Well… Now Trouble StartsEven if we simplify and assume that 1 size unit = 1 yellow unit how do we find the optimum?
Well… Now Trouble StartsEven if we simplify and assume that 1 size unit = 1 yellow unit how do we find the optimum?Even worse: the Blue Jay’s perception of size and yellow might even be negatively correlated! i.e. if sizes are hard to distinguish then yellows are not and conversely.
Well… Now Trouble StartsEven if we simplify and assume that 1 size unit = 1 yellow unit how do we find the optimum?Even worse: the Blue Jay’s perception of size and yellow might even be negatively correlated! i.e. if sizes are hard to distinguish then yellows are not and conversely.Luckily this can be dealt with.
Well… Now Trouble StartsEven if we simplify and assume that 1 size unit = 1 yellow unit how do we find the optimum?Even worse: the Blue Jay’s perception of size and yellow might even be negatively correlated! i.e. if sizes are hard to distinguish then yellows are not and conversely.Luckily this can be dealt with.For now we assume equal weights and independence.
The New ProblemHere each is a side channel curve with 2 samples
Power Sample 2
PowerSample 1
Power
Time
Note: easy generalization to more properties (will be see later)
Find the smallest sphere containing all colors
Interesting Propertymin(Problem) = min(min(Problem 1),min(Problem 2))
Hence:• Iterate the process (possibly cutting along 2 dimensions) until
halving hits a fix-point (i.e. adding compensates halving).• Solve each sub-problem by any algorithm and compute the
min over all solutions.
We Are Now Ready to
• Settle the question of dimension units.• Take correlations into account.• Justify Euclidean distance.
How to Evaluate a Distance?
stand for yellow levels and for sizes
¿ 𝑦 𝑖−𝑦 𝑗∨¿
¿ 𝑠𝑖−𝑠 𝑗∨¿
No brainerclose!
How to Evaluate a Distance?
stand for yellow levels and for sizes
¿ 𝑦 𝑖−𝑦 𝑗∨¿
¿ 𝑠𝑖−𝑠 𝑗∨¿
No brainerclose!
No brainerfar!
How to Evaluate a Distance?
stand for yellow levels and for sizes
¿ 𝑦 𝑖−𝑦 𝑗∨¿
¿ 𝑠𝑖−𝑠 𝑗∨¿
No brainerclose!
No brainerfar!??
When how many yellow saturation units worth one mm of size?!
¿ 𝑦 𝑖−𝑦 𝑗∨¿
¿ 𝑠𝑖−𝑠 𝑗∨¿
No brainerclose!
No brainerfar!??
How to Evaluate a Distance?
Compare apples to tomatoes?!
¿ 𝑦 𝑖−𝑦 𝑗∨¿
¿ 𝑠𝑖−𝑠 𝑗∨¿
No brainerclose!
No brainerfar!??
How to Evaluate a Distance?
Are we done? No!
The above assumes that butterfly yellowness and size are independent.If they were not, a hidden parameter will “count twice”!Example: weight and size are not independent.
Dealing With Correlations
Assume that we want to use, size, weight and yellowness.Evidently but
This is easy to generalize to more parameters with more correlations. cf. full paper.Euclidian distance stems from assumed multi-dimensional Gaussian distributions.
ws
ijij
y
ij
s
ij
w
ijji
sswwyysswwd
)1())((2)(
)1()(
)1()(
22
2
22
2
22
2
,