L2 is preferable when there are fewer image samples as in the case for a 5-way 1-shot case (where the use of L2 leads to 5% higher accuracy). However, as the number of support images increases, L1 shows comparable results. L2 however is more robust with quantization errors as shown in Tab. III. By using a TCAM, the complexity is no longer bounded by the number of entries M , which greatly helps in the scaling the TCAM up to more complex problems.