Technical Program

Paper Detail

Paper Title Broadcasting on Random Networks
Paper IdentifierWE2.R5.1
Authors Anuran Makur, Elchanan Mossel, Yury Polyanskiy, Massachusetts Institute of Technology, United States
Session Broadcast Channels
Location Saint Victor, Level 3
Session Time Wednesday, 10 July, 11:40 - 13:20
Presentation Time Wednesday, 10 July, 11:40 - 12:00
Manuscript  Click here to download the manuscript
Abstract We study a generalization of the problem of broadcasting on trees to the setting of directed acyclic graphs (DAGs). At time $0$, a source vertex $X$ transmits a uniform bit along binary symmetric channels (BSCs) to a set of vertices called layer $1$. Each vertex except $X$ has indegree $d$. At time $k \geq 1$, vertices at layer $k$ apply $d$-input Boolean processing functions to their received bits and send out the results to vertices at layer $k+1$. We say that broadcasting is possible if we can reconstruct $X$ with probability of error bounded away from $\frac{1}{2}$ using the values of all vertices at an arbitrarily deep layer $k$. This question is closely related to models of reliable computation and storage, probabilistic cellular automata, and information flow in biological networks. In this work, we analyze randomly constructed DAGs and demonstrate that broadcasting is only possible if the BSC noise level is below a certain (degree and function dependent) critical threshold. Specifically, for every $d \geq 3$, we identify the critical threshold for random DAGs with layers of size $\Omega(\log(k))$ and majority processing functions. For $d = 2$, we establish a similar result for the NAND processing function. Furthermore, for odd $d \geq 3$, we prove that the identified thresholds cannot be improved by other processing functions if reconstruction is required from a single vertex. Finally, for any BSC noise level, in quasi-polynomial or randomized polylogarithmic time in the depth, we construct deterministic bounded degree DAGs with layers of size $\Theta(\log(k))$ that admit reconstruction using lossless expander graphs.