Ang A. Li

Ang A. Li

Changepoint detection is increasingly applied to ecological time series, yet statistical power at the short series lengths typical of monitoring (10-50 observations) is rarely assessed. We present a simulation-based power analysis for BIC-based Binary Segmentation across 108 combinations of series length, effect size, and number of changepoints. BIC achieves $\geq$80% power for a single changepoint only at $n \geq 30$ with effect size $\geq 2.0$; detecting 2-3 changepoints requires $n \geq 50$ and ES $\geq 5.0$. BIC is conservative, underestimating changepoints more often than overestimating. AR(1) autocorrelation ($ϕ= 0.6$) reduces BIC-Binseg power by 40%, but PELT with a standard penalty maintains 85-91% power even under moderate autocorrelation. Comparison with early warning signal (EWS) variance-trend tests reveals a crossover: at ES $< 1.5$, EWS outperforms changepoint detection, but EWS rates are invariant to effect size ($\sim$73%), suggesting noise detection rather than genuine signals. Cross-system empirical validation on coral reef (Moorea, $n = 18$) and desert rodent (Portal Project, $n = 49$) time series confirms that detection succeeds when effect sizes fall in the predicted "reliable" zone. We provide power heatmaps as practical lookup tools and recommend that ecologists prefer PELT over Binseg-BIC for autocorrelated data, compute expected effect sizes before applying changepoint analysis, and pair results with permutation tests.

Ang A. Li, Zongqing Lu, Chenglin Miao

Experience replay enables off-policy reinforcement learning (RL) agents to utilize past experiences to maximize the cumulative reward. Prioritized experience replay that weighs experiences by the magnitude of their temporal-difference error ($|\text{TD}|$) significantly improves the learning efficiency. But how $|\text{TD}|$ is related to the importance of experience is not well understood. We address this problem from an economic perspective, by linking $|\text{TD}|$ to value of experience, which is defined as the value added to the cumulative reward by accessing the experience. We theoretically show the value metrics of experience are upper-bounded by $|\text{TD}|$ for Q-learning. Furthermore, we successfully extend our theoretical framework to maximum-entropy RL by deriving the lower and upper bounds of these value metrics for soft Q-learning, which turn out to be the product of $|\text{TD}|$ and "on-policyness" of the experiences. Our framework links two important quantities in RL: $|\text{TD}|$ and value of experience. We empirically show that the bounds hold in practice, and experience replay using the upper bound as priority improves maximum-entropy RL in Atari games.