In the standard ball-in-bins experiment, a well-known scheme is to sample d bins independently and uniformly at random and put the ball into the least loaded bin. It can be shown that this scheme yields a maximum load of log log n/ log d + O(1) with high probability.Subsequent work analyzed the model when at each time, d bins are sampled through some correlated or non-uniform way. However, the case when the sampling for different balls are correlated are rarely investigated. In this paper we propose three schemes for the ball-in-bins allocation problem. We assume that there is an underlying k-regular graph connecting the bins. The three schemes are variants of power-of-d choices, except that the sampling of d bins at each time are based on the locations of d independently moving non-backtracking random walkers, with the positions of the random walkers being reset when certain events occurs. We show that under some conditions for the underlying graph that can be summarized as the graph having large enough girth, all three schemes can perform as well as power-of-d, so that the maximum load is bounded by log log n/ log d + O(1) with high probability.high probability, where 1 < φ d < 2 [15]. Kenthapadi and Panigraphy [9] analyzed the scheme where d = 2 bins are sampled through picking a random edge of an underlying graph G, where G is assumed to be almost n ε regular for some 0 < ε < 1. It was shown that this scheme yields a maximum load of log log n + O(1/ε) [9]. As a comparison, Azar's model corresponds to the case that G is a complete graph. Godfrey [6] then generalized the results to the case where an indefinite number of bins are sampled through randomly picking a subset B i ⊂ [n] according to some probability measure. It was shown that, when the size of the subset is approximately Θ(log n), and for each bin j ∈ [n], the probability that j ∈ B i is at the same order of (log n)/n, the maximum load is Θ(1) with high probability. A salient feature of the scheme in [6] is that the subsets of bins can be arbitrarily correlated across balls, and only when Θ(log n) bins get sampled is there sufficient spread in the selections. More specifically, the power-of-d choices scaling doesn't hold if the number of bins sampled is some finite number d. Sampling using a single random walk on a graph has also been utilized to probe bins in balls-in-bins models. Pourmiri [12] proposed and analyzed the scheme where the placement bin for ball i is sampled using minimally loaded bins from a specific subset of locations visited by a single non-backtracking random walk of length lr G = o(log(n)) (where l = o(log(n)) and r G ∼ log log(n)) on a high girth d-regular graph between the bins, which starts from a uniformly random position in the graph. It is shown for sparse d-regular graphs (d ∈ [3, O(log(n))]) with high girth that when l ≥ 4 log n/ log k, the maximum load is O(log log n/ log(l/ log n/ log k)) with high probability. Again a salient feature of [12] is the correlated sampling of bins for each balls, but with independence of th...