Xiaochen Kev Gao
Research
Teaching
Personal
CV
class ClaimGraph(nn.Module):
def forward(self, x, adj):
h = self.gnn(x, adj)
a = softmax(q @ k.T)
return self.head(a @ v)
loss = ce(model(batch), y)
val
train
tokens = tokenizer(abstract)
emb = encoder(tokens).mean(1)
for epoch in range(n_epochs):
opt.zero_grad()
loss.backward(); opt.step()
print(f'auc = {auc:.3f}')
τ(x) = E[ Y(1) − Y(0) | X = x ]
heterogeneous treatment effects
x
τ(x)
fit η̂ on k ≠ i
score fold i
E[ ψ(W; θ
0
, η
0
) ] = 0
orthogonal score · cross-fitting
S
D
P
Q
(1)
(2)
treat
0.142
0.118
(0.031)
(0.029)
controls
✓
fixed effects
✓
✓
N
48,310
48,310
fit η̂ on k ≠ i
score fold i
E[ ψ(W; θ
0
, η
0
) ] = 0
τ(x) = E[ Y(1) − Y(0) | X = x ]
heterogeneous treatment effects
x
τ(x)
S
D
P
Q
class ClaimGraph(nn.Module):
def forward(self, x, adj):
h = self.gnn(x, adj)
a = softmax(q @ k.T)
return self.head(a @ v)
loss = ce(model(batch), y)
val
train
Personal
Under construction!