Algorithm 1 Adaptive Sample Scheduling Module
1:	procedure AdaptiveTraining
2:	Initialization:
3:	$\theta \leftarrow \mathrm{RandomInit}\left(\right)$	▹ Model parameters
4:	$\mathcal{H}\leftarrow \varnothing$	▹ Hard sample repository
5:	$S\leftarrow S_{\mathrm{init}},Q\leftarrow Q_{\mathrm{init}}$
6:	$\rho \leftarrow {\displaystyle \frac{S}{S+Q}}$	▹ Initial support ratio
7:	$\mathrm{no}\_\mathrm{improve}\_\mathrm{count}\leftarrow 0$	▹ Stagnation counter
8:	${\mathcal{A}}_{\mathrm{best}}\leftarrow -\infty$	▹ Best accuracy
9:	while not converged do
10:	MiniBatchTraining($\theta$, S, Q)
11:	if $\mathrm{epoch}modT=0$ then
12:	${\mathcal{A}}_{\mathrm{val}}\leftarrow \mathrm{Evaluate}\left(\theta \right)$
13:	if ${\mathcal{A}}_{\mathrm{val}}>{\mathcal{A}}_{\mathrm{best}}$ then
14:	${\mathcal{A}}_{\mathrm{best}}\leftarrow {\mathcal{A}}_{\mathrm{val}}$
15:	$\mathrm{no}\_\mathrm{improve}\_\mathrm{count}\leftarrow 0$
16:	$\mathcal{H}\leftarrow \mathcal{H}\cup \mathrm{CollectHardSamples}\left(\right)$
17:	else
18:	$\mathrm{no}\_\mathrm{improve}\_\mathrm{count}\leftarrow \mathrm{no}\_\mathrm{improve}\_\mathrm{count}+1$
19:	if $\mathrm{no}\_\mathrm{improve}\_\mathrm{count}\ge \tau$ then
20:	$\rho \leftarrow min(\rho +{\rho }_{\mathrm{step}},{\rho }_{max})$
21:	$N\leftarrow S+Q$	▹ Maintain total samples
22:	$S\leftarrow max(\lfloor \rho N\rfloor ,1)$
23:	$Q\leftarrow max(N-S,1)$
24:	end if
25:	end if
26:	end if
27:	end while
28:	return $\theta$
29:	end procedure