SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.02111 · ICML · 2024

Accelerating Look-ahead in Bayesian Optimization: Multilevel Monte Carlo is All you Need

Maximilian Balandat, Shangda Yang, Vitaly Zankin, Stefan Scherer, Kevin Carlberg, Neil Walton, Kody Law

arXiv · PDF · Open in the Atlas

Code that ran

3 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 1 no recorded run), measured 2026-11-21; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2402.02111.json · badge: /api/ran/2402.02111.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
Shangda-Yang/MLMCBO canonical (classifier label)
classifier basis: census
3 ran · 1 no recorded run
FunctionStatusWhere it lives
compute_nmse Ran
ran; our contract draft was wrong, not the code
Shangda-Yang/MLMCBO/tutorials/error_matrics.py
code served (permissive licence) · get_code("7a4b950d0685a135")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Construct arguments for compute_nmse
# The function expects:
# - results: a tensor that has .cpu() method, used in subtraction and division
# - costs: a tensor used with torch.cumsum and torch.mean
# - reference: a tensor for subtraction
# - relative: a tensor for division
# - R: a scalar used for division

# From the function body:
# - results and reference are subtracted, so same shape
# - error is divided by relative element-wise, so relative same shape as error
# - costs is cumsum'd along dim=1 then mean'd along dim=0
# - R is used as a divisor (scalar)
# - MSE is used with torch.std, suggesting batch dimension exists
# - len(MSE) is called, so MSE is indexable

# Let's use shape (2, 4) for batch of 2, 4 dimensions
batch_size = 2
num_dims = 4

results = torch.randn(batch_size, num_dims)
reference = torch.randn(batch_size, num_dims)
relative = torch.ones(batch_size, num_dims) + 0.1  # avoid division issues
R = 5  # scalar

# costs should work with cumsum along dim=1, so shape (batch, time_steps)
costs = torch.ones(batch_size, 3)

ARGS = {
    "results": results,
    "costs": costs,
    "reference": reference,
    "relative": relative,
    "R": R
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
optimize_mlmc Ran
ran; our contract draft was wrong, not the code
Shangda-Yang/MLMCBO/mlmcbo/utils/optimize_mlmc.py
code served (permissive licence) · get_code("a95735eb0abd4d5d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

class MockAcquisitionFunction:
    """Mock acquisition function that implements sample_candidate method."""
    
    def __init__(self):
        self.call_count = 0
    
    def sample_candidate(self, level, dl, n, match=None, match_mode='point'):
        """
        Simulate sampling candidates at a given level.
        
        Args:
            level: current level
            dl: starting level
            n: number of samples
            match: match state from previous iteration
            match_mode: 'point', 'forward', or 'backward'
        
        Returns:
            zl: candidate tensor
            fl: function value tensor
            match: updated match state
        """
        # Return tensors with batch dimension
        if match_mode == 'point':
            # For point matching, return scalar-like tensors
            zl = torch.randn(1)
            fl = torch.randn(1)
        else:
            # For forward/backward, return batch of candidates
            zl = torch.randn(n, 2)
            fl = torch.randn(n)
        
        # Update match state (can be anything, just pass it through)
        match = level
        
        return zl, fl, match

inc_function = MockAcquisitionFunction()
eps = 0.1
dl = 0
alpha = 1.0
beta = 1.5
gamma = 1.0
meanc = 1.0
varc = 1.0
var0 = 1.0
match_mode = 'point'

ARGS = {
    'inc_function': inc_function,
    'eps': eps,
    'dl': dl,
    'alpha': alpha,
    'beta': beta,
    'gamma': gamma,
    'meanc': meanc,
    'varc': varc,
    'var0': var0,
    'match_mode': match_mode,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
optimize_mlmc_three Ran
ran; our contract draft was wrong, not the code
Shangda-Yang/MLMCBO/mlmcbo/utils/optimize_mlmc.py
code served (permissive licence) · get_code("e6e12296737226de")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

class MockIncFunction:
    """Mock inc_function that provides sample_candidate method."""
    
    def __init__(self):
        self.call_count = 0
    
    def sample_candidate(self, level, dl, n, match=None, match_mode='point'):
        """
        Returns:
            zl: candidate solution tensor
            fl: acquisition function value tensor
            match: match object (can be None or any object)
        """
        # Create small tensors for batch_size=2, feature_dim=4
        batch_size = 2
        feature_dim = 4
        
        if match_mode == 'point':
            # For point mode, return scalars
            zl = torch.randn(batch_size, feature_dim)
            fl = torch.randn(batch_size)
        else:
            # For forward/backward modes, return batched tensors
            zl = torch.randn(batch_size, feature_dim)
            fl = torch.randn(batch_size)
        
        # match can be any object, just pass it through or create new
        new_match = {"level": level, "n": n}
        
        self.call_count += 1
        return zl, fl, new_match


ARGS = {
    "inc_function": MockIncFunction(),
    "eps": 0.1,
    "dl": 0,
    "alpha": 1.0,
    "beta": 1.5,
    "gamma": 1.0,
    "meanc": 1.0,
    "varc": 1.0,
    "var0": 1.0,
    "match_mode": "point",
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
dist_matrix Not yet run
unverified (harvested, no recorded run)
Shangda-Yang/MLMCBO/mlmcbo/acquisition_functions/mlmc_inc_functions.py
code served (permissive licence) · get_code("5821b16230a57328")
How it ran

no invocation recorded.

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 4 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-21.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We leverage multilevel Monte Carlo (MLMC) to improve the performance of multi-step look-ahead Bayesian optimization (BO) methods that involve nested expectations and maximizations. Often these expectations must be computed by Monte Carlo (MC). The complexity rate of naive MC degrades for nested operations, whereas MLMC is capable of achieving the canonical MC convergence rate for this type of problem, independently of dimension and without any smoothness assumptions. Our theoretical study focuses on the approximation improvements for twoand three-step look-ahead acquisition functions, but, as we discuss, the approach is generalizable in various ways, including beyond the context of BO. Our findings are verified numerically and the benefits of MLMC for BO are illustrated on several benchmark examples. Code is available at https://github.com/Shangda-Yang/MLMCBO.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2402.02111")
get_code_for_paper("2402.02111")
have("2402.02111")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2402.02111.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2402.02111.svg)](https://syntology.ai/paper/2402.02111)

Connect an agent — have() is free.