Beepul Bharti, Jacopo Teneggi, Yaniv Romano, Jeremias Sulam
Explainable AI Seminars @ Imperial
5/30/2024
[M. E. Kaminski, 2019]
E.U.: “right to an explanation” of decisions made
on individuals by algorithms
F.D.A.: “interpretability of the model outputs”
efficiency
nullity
symmetry
exponential complexity
Lloyd S Shapley. A value for n-person games. Contributions to the Theory of Games, 2(28):307–317, 1953.
Let be an - person cooperative game with characteristic function
How important is each player for the outcome of the game?
inputs
responses
predictor
inputs
responses
predictor
inputs
responses
predictor
What does it mean for a feature to receive a high Shapley Value?
Precise Notions of Importance
[Candes et al, 2018]
Do the features in \(S\) contain any additional information about the label \(Y\) that is not in the complement \({[n]\setminus S}\) ?
Precise Notions of Importance
XRT: eXplanation Randomization Test
returns a valid p-value, \(\hat{p}_{i,S}\) for the test above
Precise Notions of Importance
Given the Shapley coefficient of any feature
Then
and the (expected) p-value obtained for , i.e.
Theorem 1:
Teneggi, Bharti, Romano, and S. "SHAP-XRT: The Shapley Value Meets Conditional Independence Testing." TMLR (2023).
What does the Shapley value test?
Given the Shapley value for the i-th feature, and
Theorem 2:
Then, under \(\displaystyle H^0_\text{global} \), \(p_{\text{global}}\) is a valid p-value and
Full Spectrum of Tests
S
S
\(\hat{p}_{4,S}\)
\(\hat{p}_{2,S}\)
Input features are not inherently
interpretable to users
So far
model
\[\hat{y} = f(x)\]
statistical importance of
\[x_j\]
input features \(x_j\)
\(\rightarrow\)
concepts \(c_j\)
input features \(x_j\)
\(\rightarrow\)
concepts \(c_j\)
What are the important concepts for
the predictions of a model?
Global and local importance
Fixed predictor
Statistically-valid importance
Any set of concepts
\(x \in \mathcal{X}\)
\(f:~\mathcal{X} \to \mathbb{R}^d\)
\(g:~\mathbb{R}^d \to [0, 1]^k\)
\(c \in \mathbb{R}^{d \times m}\)
input
encoder
classifier
concepts
\(Z = c^{\top} H\)
semantics
\(\hat{Y} = g(f(X))\)
predictions
\(H = f(X)\)
embeddings
\(x \in \mathcal{X}\)
\(f:~\mathcal{X} \to \mathbb{R}^d\)
\(g:~\mathbb{R}^d \to [0, 1]^k\)
\(c \in \mathbb{R}^{d \times m}\)
input
encoder
classifier
concepts
\(Z = c^{\top} H\)
semantics
\(\hat{Y} = g(f(X))\)
predictions
\(H = f(X)\)
embeddings
FIXED
For a concept \(j \in [m] \coloneqq \{1, \dots, m\}\)
type of dependence
marginal
over a population
global importance
\[H^{\text{G}}_{0,j}:~\hat{Y} \perp \!\!\! \perp Z_j\]
\(Z_j\)
\(\hat{Y}\)
For a concept \(j \in [m] \coloneqq \{1, \dots, m\}\)
type of dependence
conditional
over a population
global conditional importance
\[H^{\text{GC}}_{0,j}:~\hat{Y} \perp \!\!\! \perp Z_j \mid Z_{-j}\]
\(Z_j\)
\(\hat{Y}\)
\(Z_{-j}\)
For a concept \(j \in [m] \coloneqq \{1, \dots, m\}\)
type of dependence
conditional
for a specific input
semantic XRT
\[H^{\text{LC}}_{0,j,S}:~g(\widetilde{H}_{S \cup \{j\}}) \overset{d}{=} g(\widetilde{H}_S),\quad\widetilde{H}_C \sim P_{H | Z_C = z_C}\]
\(z_j\)
\(\hat{Y}\)
\(z_S\)
Null hypothesis \(H_0\), significance level \(\alpha \in (0,1)\)
Classical approach
collect data
analyze data
compute \(p\)-value
\(\rightarrow\)
\(\rightarrow\)
"Reject \(H_0\) if \(p \leq \alpha\)"
Sequential approach
Instantiate wealth process \(K_0 = 1\), \(K_t = K_{t-1} \cdot (1 + v_t\kappa_t)\)
"Reject \(H_0\) when \(K_t \geq 1/\alpha\)"
time
wealth
\(1/\alpha\)
under \(H_0\)
under \(H_1\)
rejection time \(\tau\)
Data-efficient
Only use the data is needed to reject
Adaptive
The harder to reject, the longer the test
\[\downarrow\]
If concept \(c\) rejects faster than \(c'\)
then it is more important
Induces a natural rank of importance
All definitions of semantic importance are two-sample tests
\[H_0:~P = \widetilde{P}\]
Global importance (SKIT)
\(H^{\text{G}}_{0,j}\)
\(P_{\hat{Y} Z_j} = P_{\hat{Y}} \times P_{Z_j}\)
\(\iff\)
\(H^{\text{GC}}_{0,j}\)
Global conditional importance (c-SKIT)
\(\iff\)
\(P_{\hat{Y} Z_j Z_{-j}} = P_{\hat{Y} \widetilde{Z}_j, Z_{-j}},~\widetilde{Z}_j \sim P_{Z_j | Z_{-j}}\)
\(H^{\text{LC}}_{0,j,S}\)
Local conditional importance (x-SKIT)
\(\iff\)
\(P_{\hat{Y}_{S \cup \{j\}}} = P_{\hat{Y}_S} \)
All definitions of semantic importance are two-sample tests
\[H_0:~P = \widetilde{P}\]
Global importance (SKIT)
\(H^{\text{G}}_{0,j}\)
\(P_{\hat{Y} Z_j} = P_{\hat{Y}} \times P_{Z_j}\)
\(\iff\)
\(H^{\text{GC}}_{0,j}\)
Global conditional importance (c-SKIT)
\(\iff\)
\(P_{\hat{Y} Z_j Z_{-j}} = P_{\hat{Y} \widetilde{Z}_j, Z_{-j}},~\widetilde{Z}_j \sim P_{Z_j | Z_{-j}}\)
\(H^{\text{LC}}_{0,j,S}\)
Local conditional importance (x-SKIT)
\(\iff\)
\(P_{\hat{Y}_{S \cup \{j\}}} = P_{\hat{Y}_S} \)
novel procedures
global importance
global conditional importance
local conditional importance
Jeremias Sulam
Jacopo Teneggi
Yaniv Romano
Beepul Bharti
SHAP-XRT
IBYDMT