--- title: 'Empirical Success Rule: Theory & Applications' url: https://www.emergentmind.com/topics/empirical-success-rule type: topic --- # Empirical Success Rule: Theory & Applications Searching arXiv for recent and relevant papers on "empirical success rule" and related "success rule" usage. In the cited literature, the empirical success rule denotes a family of data-dependent decision procedures that act on observed success rather than on oracle knowledge or rigid significance thresholds. In statistical decision theory, it prescribes the treatment with the highest observed average patient outcome in a randomized trial; in evolutionary computation, closely related success rules adapt step size or mutation rate according to observed improvement frequencies; and in sequential stopping with unknown Bernoulli parameter, a plug-in rule replaces the unknown success probability by an online empirical estimate [2006.00343] [1802.03209] [2604.07183]. ## 1. Formal statistical-decision meaning Manski and Tetenov formulate the empirical success rule as a treatment rule for randomized trials. For treatments \(t=1,\ldots,L\), let \[ \hat{u}_t=\frac{1}{n_t}\sum_{i:\, t_i=t}u(y_i), \] where \(u(\cdot)\) is the patient-relevant outcome function. The rule chooses \[ t^*=\arg\max_t \hat{u}_t, \] and, if there is a tie, randomizes evenly among maximizers [2006.00343]. A central property of this formulation is symmetry across trial arms. The rule does not defer to standard care as a status quo and does not require that an innovation pass a statistical-significance threshold before it can be chosen. In the COVID-19 trial setting studied by Manski and Tetenov, this symmetry is presented as especially relevant because standard care itself may be uncertain or suboptimal [2006.00343]. Evaluation is conducted through near-optimality, defined via maximum regret. In state \(s\), \[ \text{Regret in state } s = \max_t u_{st} - \sum_t u_{st}\,\mathbb{E}_s[\delta(t,\psi)], \] where \(u_{st}\) is the true mean outcome of treatment \(t\), \(\mathbb{E}_s[\delta(t,\psi)]\) is the expected fraction assigned to \(t\) by rule \(\delta\), and \(\psi\) is the observed data. Near-optimality is the maximum regret over all possible states. This criterion jointly accounts for the probability and magnitude of decision errors [2006.00343]. ## 2. Near-optimality and comparison with hypothesis testing In the trial framework, the empirical success rule is compared with the conventional practice of choosing an innovation only if the estimated treatment effect is positive and statistically significant. Manski and Tetenov argue that the conventional rule is asymmetric, defers to standard care, and can incur substantial type II error costs when trials are small and estimates are imprecise [2006.00343]. The reported quantitative comparisons are sharp. In two-arm trials with 100 patients per group and binary outcome, the empirical success rule achieves near-optimality of \(0.012\), versus \(0.071\) for \(5\%\) hypothesis testing; the paper states that the empirical success rule is about six times nearer to optimality. In multi-arm trials, including five-arm settings, the regret of the empirical success rule is reported as more than three times lower than that of multiple-hypothesis-test-based rules. The same summary states that using the empirical success rule with 100 patients per arm confers the same reduction in maximum expected loss as massively increasing sample size for hypothesis testing, with the approximation “ESR with 100/arm \(\approx\) Hypothesis testing with 4000/arm” [2006.00343]. The illustrative example given in the same work emphasizes magnitude-sensitive decision error. When mortality under a new treatment is \(0.15\) and under standard care is \(0.25\), the hypothesis-test rule fails to adopt the better treatment with probability \(57\%\), yielding expected loss \(0.057\); the empirical success rule yields expected loss \(0.004\) [2006.00343]. The paper further reports that the empirical success rule achieves the minimal possible near-optimality in binary, balanced trials, citing Stoye and Hirano & Porter. It also states that the good properties extend to multi-arm trials, to heterogeneous patients by applying the rule within covariate strata, and to multiple outcomes by defining composite welfare measures [2006.00343]. ## 3. Success-based adaptation in evolutionary computation In evolutionary computation, the most prominent related construct is the one-fifth success rule originating with Rechenberg. In the \((1+1)\)-Evolution Strategy analyzed on the sphere function, the algorithm maintains a mean vector \(m_t\) and a step size \(\sigma_t\). Each iteration samples \[ x_t \sim N(m_t,\sigma_t^2 I). \] If \(f(x_t)\le f(m_t)\), the step is a success, \(m_{t+1}\leftarrow x_t\), and the step size is increased according to \(\sigma_{t+1}\leftarrow \sigma_t\cdot \alpha\). Otherwise, \(m_{t+1}\leftarrow m_t\) and \(\sigma_{t+1}\leftarrow \sigma_t\cdot \alpha^{-1/4}\). The supplied summary states that this empirical success rule directly adapts the step size to maintain a success probability near \(1/5\), and that the normalized step size governs the probability of success [1802.03209]. The same idea has been generalized in discrete evolutionary algorithms as a family of success-based update rules parameterized by an update strength \(F>1\) and a success ratio \(s>0\). In the self-adjusting \((1+1)\) EA on LeadingOnes, a successful offspring updates the mutation rate to \(F^s\rho\), while a failure updates it to \(\rho/F\), with clipping to \(\rho_{\max}\) and \(\rho_{\min}\). The generalized target success rate is \(1/(s+1)\), so the classical one-fifth rule corresponds to \(s=4\) [1902.02588]. A misconception addressed by this line of work is that the literal one-fifth target is universally optimal. For LeadingOnes, the best performance is reported for small update strengths \(F=1+o(1)\) and success rate \(1/e\), rather than \(1/5\). At fitness level \(\ell\), the target mutation rate is defined by \[ (1-\rho^*)^{\ell}=\frac{1}{s+1}, \] which yields \[ \rho^*(\ell,s)=1-(s+1)^{-1/\ell}\approx \frac{\ln(s+1)}{\ell}. \] The main runtime bound is \[ E[T]\le (1+o(1))\frac{s+1}{4\ln(s+1)}n^2, \] and the minimum occurs at \(s=e-1\), corresponding to the \(1/e\) target success rate [1902.02588]. ## 4. Drift theory, unbounded potentials, and continuous-domain hitting times The 2018 analysis of the \((1+1)\)-ES with one-fifth success rule establishes the first explicit application of drift analysis to a randomized search heuristic with continuous domain, and the first non-asymptotic analysis for that algorithm. The technical obstacle is that natural potentials such as \(\log(\|m_t\|)\) are unbounded below as the optimum is approached, and rare but extremely good steps can invalidate simple arguments unless controlled [1802.03209]. To address this, the paper introduces new drift theorems for unbounded-below potential functions. The upper-bound theorem uses a truncated process \[ Y_{t+1}=Y_t+\max\{X_{t+1}-X_t,-A\}, \] and if the conditional expected truncated drift satisfies \[ \mathbb{E}[Y_{t+1}\mid \mathcal{F}_t]-Y_t\le -B, \] then for the hitting time \(T\) of target \(a\), \[ \mathbb{E}[T]\le \frac{x_0-a+A}{B}. \] A corresponding lower-bound theorem is given for monotone non-increasing and integrable processes with expected decrease at least \(-C\), yielding \[ \mathbb{E}[T]\ge \frac{x_0-a}{4C}-\frac12. \] These theorems permit drift analysis with potentials such as \(\log(\|m_t\|)\) [1802.03209]. The constructed potential couples optimization and adaptation: \[ V(m,\sigma)=\log(\|m\|)+\max\left\{0,\; v\log\left(\frac{\alpha \ell \|m\|}{d\sigma}\right),\; v\log\left(\frac{\alpha^{1/4}\sigma d}{u\|m\|}\right)\right\}, \] where \(v>0\), and \([\ell,u]\) defines a well-adapted range for the normalized step size. The first term tracks optimization progress; the max-terms penalize step sizes that are too small or too large, thereby forcing adaptation [1802.03209]. Let \[ T_\epsilon=\min\{t:\|m_t\|\le \epsilon\}. \] The paper proves \[ \frac{(\log(\|m_0\|)-\log(\epsilon))\,d}{4}-\frac12 \;\le\; \mathbb{E}[T_\epsilon] \;\le\; \frac{V(\theta_0)-\log(\epsilon)+\frac1d}{B}, \] with \(B\in \Theta(1/d)\). As summarized in the supplied notes, \[ \mathbb{E}[T_\epsilon]\in \Theta\!\left((\log(\|m_0\|)+\log(1/\epsilon))\cdot d\right), \] so the convergence rate scales as \(\Theta(1/d)\). A matching lower-bound ingredient is that even hit-and-run does not allow more than \[ \mathbb{E}[\log(\|m_t\|)-\log(\|m_{t+1}\|)\mid \mathcal{F}_t]\le \frac1d \] per iteration [1802.03209]. ## 5. Discrete mutation control and hybrid learning-based rules The discrete theory sharpens the interpretation of success rules as online parameter control. For LeadingOnes, the self-adjusting \((1+1)\) EA with small \(F\) and optimal success ratio \(s=e-1\) attains \[ E[T]\approx 0.68 n^2 + o(n^2), \] which the supplied summary states matches the asymptotic runtime of the \((1+1)\) EA with the best possible fitness-dependent mutation rate. The one-fifth rule remains competitive but slightly worse, at approximately \(0.777n^2\), while still outperforming any static mutation rate. The same summary reports that for \(s\in[0.78,3.92]\), the self-adjusting EA outperforms any static mutation rate [1902.02588]. This result clarifies a second misconception: success rules are not merely heuristic surrogates for finely tuned schedules. Under the analyzed conditions, they can be asymptotically equivalent to an oracle fitness-dependent mutation strategy. The paper also states that similar results hold for the resampling variant of the \((1+1)\) EA, which enforces at least one bit flip per iteration [1902.02588]. A further extension combines success-based control with reinforcement learning. The hybrid Q-learning mechanism HQL controls the mutation rate of the \((1+\lambda)\) EA by maintaining a Q-table over states defined by the number of successful offspring in a generation and over two actions: multiplying the current mutation rate by \(A\) or decreasing it by \(b\). If the Q-values of the two actions are equal, the action-selection tie is broken by the one-fifth-th success rule: increase the mutation rate if the best offspring is strictly better than the parent, and decrease it otherwise [2006.11026]. The reported empirical finding is that HQL achieves equal or superior performance to all techniques tested in Rodionova et al. and does so simultaneously for all offspring population sizes \(\lambda\). The same source states that the promising performance is not restricted to OneMax but extends to several other benchmark problems, including LeadingOnes, neutrality, plateau, and ruggedness functions. It also notes that no full theoretical convergence analysis is provided because tracking Q-values and state dependencies is complex [2006.11026]. ## 6. Plug-in stopping rules, regulatory usage, and limits of oracle-freeness A structurally related use of empirical success appears in the last-success problem with unknown Bernoulli parameter. With \(X_1,\ldots,X_n\) i.i.d. \(\mathrm{Bernoulli}(p)\) and \(p\) unknown, the plug-in rule estimates \[ \hat p_t=\frac{S_t}{t},\qquad S_t=\sum_{i=1}^t X_i, \] and stops at \[ \hat{\tau}_n = \inf\left\{ t\in\{1,\ldots,n\}: X_t=1 \text{ and } \left(\hat p_t<\frac{1}{n-t+1}\ \text{or } t=n\right) \right\}. \] Its win probability is \[ W_n(p)=\mathbb{P}_p\!\left[\hat\tau_n\le n,\ X_{\hat\tau_n}=1,\ X_{\hat\tau_n+1}=\cdots=X_n=0\right]. \] The paper derives an exact recursion for state probabilities \(u_{t,k}\) and a representation of \(W_n(p)\) that can be computed in \(O(n^2)\) time [2604.07183]. The same work proves that for \(n\ge 6\), the plug-in rule is always suboptimal in finite horizon: \[ W_n(p)