1: \begin{abstract}

2: An information theoretic perspective on group testing problems has

3: recently been proposed by Atia and Saligrama, in order to characterise

4: the optimal number of tests. Their results hold in the noiseless case,

5: where only false positives occur, and where only false negatives occur.

6: We extend their results to a model containing both false positives

7: and false negatives, developing simple information theoretic bounds

8: on the number of tests required. Based on these bounds, we obtain

9: an improved order of convergence in the case of false negatives only.

10: Since these results are based on (computationally infeasible) joint

11: typicality decoding, we propose a belief propagation algorithm for

12: the detection of defective items and compare its actual performance

13: to the theoretical bounds. 

14: \end{abstract}