Diversity Learning: Introducing the Space-time Scheme to Ensemble Learning

Statistical Tools and Methodologies for Ultrareliable Low-Latency Communication—A Tutorial

“…1) Explainable ML [42], [43] facilitates understanding the ML decision-making processes, allowing for increased transparency and trust in the algorithms. 2) Ensemble learning [44], [45] involves combining multiple ML models to improve the accuracy and robustness of the predictions. 3) Transfer learning [32], [46], FL [38], and metalearning [47] are specially efficient in low-data scenarios.…”

“…Consequently, the probability estimate of F 1 is Pr[F 1 ] = q, which can be estimated via CMC, since F 1 is by design not rare, and the number of samples, n, is small. Moreover, we can calculate the rare-event probability as in (45), which for the first domain reduces to q Pr[F |F 1 ]. Therefore, Pr[F |F 1 ] uses a Markov-chain MC method (line 7), e.g., a modified MH or Gibbs sampling [98], [99].…”

“…This process is repeated iteratively until the stopping criterion is met (see [99] for details on the stopping criteria and an alternative implementation of the algorithm). By design, the SS algorithm controls the probability of the intermediate steps; therefore, (45) approximates to q L Pr[F |F l ], where l ≤ L indicates the number of domains, Pr[F |F l ] = n(L)/n, and n(L) is the number of samples in the last domain. Thus, the output of line 8 in Algorithm 2 is the rare-event probability, which is given as…”

“…A key advantage of the SS algorithm is that by design 1, can control input variable q, which influences the number of chains, the number of samples in each chain, and the stopping criteria. Therefore, with proper selection of q, the SS algorithm efficiently samples the conditional distributions to (approximately) match this probability ensuring that each factor in (45) is larger than q. Such an iterative process yields an accurate estimate of the rare event even for a small sample size.…”