Correction to: Rule-Based Runtime Mitigation Against Poison Attacks on Neural Networks

Correction to: Chapter “Rule-Based Runtime Mitigation Against Poison Attacks on Neural Networks” in: T. Dang and V. Stolz (Eds.): Runtime Verification, LNCS 13498, https://doi.org/10.1007/978-3-031-17196-3_4

In an older version of this paper, there was error in the figure 3, (e) and (f) was incorrect. This has been corrected.

Fig. 3.

Rule-based detection and correction of poison attacks. Rule(1) =$N1 < A_{1}$. Rule(2) =$(N_{1} \ge A_{1}) \wedge (N_{1} < A_{3}) \wedge (N_{0} \ge B_{2})$. Rule(3) =$(N_{1} \ge A_{3}) \wedge (N_{0} \ge B_{1})$. Rule(4) =$(N_{1} \ge A_{1}) \wedge (N_{1}< A_{2}) \wedge (N_{0} < B_{2})$. Rule(5) =$(N_{1} \ge A_{2}) \wedge (N_{1}< A_{3}) \wedge (N_{0} \ge B_{1}) \wedge (N_{0} < B_{2})$. Rule(6) =$(N_{1} \ge A_{2}) \wedge (N_{0} < B_{1})$. Rule(7) =$(N_{1} \ge A_{5}) \wedge (N_{1}< A_{6}) \wedge (N_{0} \ge B_{5}) \wedge (N_{0} < B_{6})$. Rule(1$'$) =$N1 < A_{1}' \wedge (N_{0} \ge B_{2})$. Rule(2$'$) =$(N_{1} \ge A_{1}') \wedge (N_{1} < A_{3}) \wedge (N_{0} \ge B_{2})$. Rule(4$'$) =$(N_{1}< A_{2}) \wedge (N_{0} < B_{2})$. $A_{i}$ and $B_i$ are threshold values for $N_0$ and $N_1$ respy. (Color figure online)