{
 "artifacts": null,
 "category": "sampling",
 "description": "Train a cheap neural surrogate globally, then use an ensemble or bootstrap covariance to identify inputs near the estimated upper-tail boundary and inputs where high-fidelity correction is uncertain. Fit a Tikhonov-regularized residual model on the acquired expensive labels and use the corrected predictor for CVaR estimation or risk-constrained optimization. The acquisition policy deliberately ignores easy central-region samples unless they influence the tail threshold.",
 "formulas_latex": [
  "$$\\mathrm{VaR}_{\\beta;\\mathbf{d}}(y(\\mathbf{X})):=\\inf\\{t\\in\\mathbb{R}:\\mathrm{P}_{\\mathbf{d}}(y(\\mathbf{X})\\leq t)\\geq\\beta\\}.$$",
  "$$\\mathrm{CVaR}_{\\beta}(Y)=\\min_{\\eta\\in\\mathbb{R}}\\left[\\eta+\\frac{1}{1-\\beta}\\mathbb{E}\\left[(Y-\\eta)_{+}\\right]\\right],\\qquad (a)_{+}=\\max(a,0).$$",
  "$$\\mathcal{R}_{\\epsilon}=\\left\\{x:\\left|\\mu_L(x)-q_{\\beta}^{L}\\right|\\leq\\epsilon s_L(x)\\right\\},\\qquad \\widehat{y}(x)=\\mu_L(x)+\\phi(x)^{\\top}\\widehat{w}.$$",
  "$$\\widehat{w}=\\arg\\min_{w}\\sum_{i\\in H}\\left(y_{H,i}-y_{L,i}-\\phi(x_i)^{\\top}w\\right)^2+\\lambda\\|w\\|_2^2,\\quad a_1(x)=s_L(x),\\quad a_2(x)=s_r(x)+\\rho\\left|\\phi(x)^{\\top}\\widehat{w}\\right|.$$"
 ],
 "id": 2816,
 "implementation": "(1) Exact integration point: place this module around a costly simulator, human labeler, safety checker, or high-resolution world model. The neural low-fidelity predictor receives candidate states or designs $x$ and produces $K$ ensemble predictions. The high-fidelity oracle is queried only for selected candidates. Use the corrected predictor in the CVaR loss, safety constraint, or downstream design optimizer; it need not replace the ordinary forward pass everywhere. (2) Pseudocode: initialize a low-fidelity model $f_L$ and a small high-fidelity set $H$; train $K$ independently initialized or bootstrapped copies; on a large cheap candidate pool compute $\\mu_L(x)=K^{-1}\\sum_k f_L^{(k)}(x)$ and $s_L^2(x)=(K-1)^{-1}\\sum_k(f_L^{(k)}(x)-\\mu_L(x))^2$; compute the empirical quantile $q_\\beta^L$; retain candidates satisfying $|\\mu_L(x)-q_\\beta^L|\\leq\\epsilon s_L(x)$; select one batch fraction by largest $s_L$ and the remainder by largest $s_r+\\rho|\\phi(x)^T\\widehat w|$; query the expensive oracle at those points; append $(x,y_H)$ to $H$; fit $\\widehat w$ using the ridge objective above; optionally fine-tune $f_L$ and repeat. Estimate CVaR on a fresh Monte Carlo pool by evaluating corrected predictions and minimizing the scalar empirical objective over $\\eta$. (3) Computed from the paper's mathematics: the empirical quantile, confidence tail region, ridge residual coefficients, corrected predictions, and CVaR objective. Estimated empirically: ensemble uncertainty $s_L$, residual uncertainty $s_r$ from bootstrap residual fits, and the candidate-pool distribution. Use standardized scores or rank-based selection so uncertainty and correction magnitude are comparable. (4) First cheap experiment: construct a 2D or 5D synthetic simulator with a cheap biased function and an expensive function containing a narrow failure region; compare random sampling, global ensemble-uncertainty sampling, and this two-stage policy at $\\beta=0.99$. Use a five-member 3-layer MLP ensemble, 32--128 initial high-fidelity labels, and batches of 16 oracle queries. Measure absolute CVaR error, high-fidelity calls required to certify a CVaR constraint, and global MSE. The method succeeds if it reaches lower CVaR error or reliable constraint satisfaction with fewer expensive evaluations than both baselines; a useful target is at least 2x fewer high-fidelity labels than random or global uncertainty sampling, even if central-region MSE is unchanged.",
 "math_summary": "The paper defines the upper-tail quantile as $\\mathrm{VaR}_{\\beta;\\mathbf{d}}(y(\\mathbf{X})):=\\inf\\{t\\in\\mathbb{R}:\\mathrm{P}_{\\mathbf{d}}(y(\\mathbf{X})\\leq t)\\geq\\beta\\}$, where $\\beta\\in(0,1)$ is the risk level, $\\mathbf{d}$ is a design or conditioning variable, $\\mathbf{X}$ is the random input, and $y$ is the response. Use the equivalent Rockafellar representation $\\mathrm{CVaR}_{\\beta}(Y)=\\min_{\\eta\\in\\mathbb{R}}[\\eta+(1-\\beta)^{-1}\\mathbb{E}(Y-\\eta)_{+}]$, with $(a)_{+}=\\max(a,0)$ and $Y=y(\\mathbf{X})$. Let $\\mu_L(x)$ and $s_L(x)$ be the mean and standard deviation of an ensemble low-fidelity predictor, and let $q_{\\beta}^{L}$ be its empirical $\\beta$-quantile over a candidate pool. Define the confidence-based tail region $\\mathcal{R}_{\\epsilon}=\\{x:\\left|\\mu_L(x)-q_{\\beta}^{L}\\right|\\leq\\epsilon s_L(x)\\}$, where $\\epsilon\u003e0$ controls the confidence band. Use a residual expansion $r(x)=\\phi(x)^{\\top}w$ fitted by Tikhonov regularization, with $H$ the high-fidelity dataset, $\\phi(x)$ a fixed feature vector, and $\\lambda\u003e0$: $\\widehat{w}=\\arg\\min_w\\sum_{i\\in H}(y_{H,i}-y_{L,i}-\\phi(x_i)^{\\top}w)^2+\\lambda\\|w\\|_2^2$. The corrected predictor is $\\widehat y(x)=\\mu_L(x)+\\phi(x)^{\\top}\\widehat w$. A two-stage acquisition policy first selects uncertain points in $\\mathcal R_\\epsilon$ using $a_1(x)=s_L(x)$, then exploits locations with large correction impact using $a_2(x)=s_r(x)+\\rho|\\phi(x)^{\\top}\\widehat w|$, where $s_r(x)$ is residual uncertainty and $\\rho\\geq0$ controls exploitation.",
 "math_tags": [
  "probability",
  "statistics",
  "optimization",
  "regularization"
 ],
 "ml_areas": [
  "loss",
  "sampling",
  "training",
  "world-model"
 ],
 "paper": {
  "arxiv_id": "2608.29222",
  "arxiv_url": "https://arxiv.org/abs/2608.29222",
  "summary_what_math_gives_to_ml": "The paper's transferable contribution is a tail-focused multifidelity correction strategy: use a cheap global predictor, quantify its finite-sample uncertainty, and spend expensive evaluations only near the estimated risk boundary and where the correction can materially change CVaR. This is more targeted than ordinary uncertainty sampling because it optimizes the quantity that matters—an upper quantile or tail average—rather than reducing average prediction error. A neural implementation can use a low-cost surrogate or frozen backbone, an uncertainty estimator such as an ensemble, and a regularized residual model trained only on high-fidelity discrepancies. The strongest application is expensive simulator or safety-label learning, where reducing error in rare catastrophic outcomes matters more than improving central-region MSE.",
  "title": "Risk-averse design optimization with CVaR constraints via multifidelity tail-region correction",
  "year": "2026"
 },
 "ratings": {
  "difficulty": 5,
  "novelty": 6,
  "usefulness": 7
 },
 "solves": [
  "sample-efficiency",
  "accuracy",
  "scalability"
 ],
 "title": "CVaR-tail active residual correction",
 "url": "https://synthcore.org/idea/2816/cvar-tail-active-residual-correction",
 "verification": {
  "peer_reviewed": false,
  "status": "unverified",
  "status_label": "Unverified",
  "verdict_source": "deterministic test code (paired-seed permutation statistics)"
 }
}