Figure 14From: Learning without lossSame as Fig. 13 but with a varying Δ and fixing the exempted data at 0.5%Back to article page