Skip to Content
All Research

Research

Improved Algorithms for Maximal Clique Search in Uncertain Networks

Li, Dai, Wang, Ming, Qin & Yu, IEEE ICDE 2019

Input
An uncertain graph G = (V, E, p), where every edge has a probability of actually existing (e.g. inferred confidence in a PPI network, or co-authorship strength in DBLP), plus two parameters: a minimum clique size k and a minimum clique probability threshold tau.
Supervision Required
None. This is a deterministic combinatorial-algorithms paper, not a learned model. The edge probabilities are given input data, not learned from anything; there's no training or supervision involved at all.
Representation Learned
Nothing learned. The paper's actual contribution is algorithmic: a new (Top_k, tau)-core pruning technique plus a cut-based optimization that shrink the search space before an improved backtracking enumeration, reducing worst-case time complexity from 2^n to 2^n' where n' is the size of a much smaller pruned subgraph.
How It Connects to Robot Actions
None. This is a graph-algorithms / database-systems project (FSU COP 5725), unrelated to robot actions or Physical AI. Included here as real algorithmic work I implemented, not because it fits the Physical AI thread. The point of this template was never to pretend every paper connects to robotics.
Real Hardware Test
Not applicable. Evaluated on six real-world graph datasets (including DBLP co-authorship and PPI protein-interaction networks) as a software benchmark, with a case study on detecting protein complexes. No hardware or embodied evaluation of any kind.
Limitations / Failure Modes
Still worst-case exponential (2^n') even after pruning. The speedup is entirely dependent on how much the (Top_k, tau)-core pruning shrinks the graph, which in turn depends on the graph's degeneracy. Real-world graphs with high degeneracy would see much less benefit from this technique.
Experiment I Would Add
Apply the same core-pruning idea to an uncertain graph outside the social/PPI-network setting the paper evaluates (e.g. a sensor-network or multi-robot connectivity graph with uncertain links) to see whether the pruning assumptions (low degeneracy) still hold in that domain.