SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.25663 · 2026

Opportunistic Target Selection: Early Directional Commitment for Query-Efficient Black-Box Adversarial Attacks

Florian Yger, Florent Tariolle

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 23 functions out of this paper's own repositories and ran 20 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
Tariolle/opportunistic-target-selection canonical 20 of 23
FunctionStatusWhere it lives
analyze_overlap Ran Tariolle/opportunistic-target-selection/analysis/analyze_target_overlap.py
code served (permissive licence) · get_code("195dacdd23c77447")
bootstrap_cdf Ran Tariolle/opportunistic-target-selection/analysis/analyze_winrate.py
code served (permissive licence) · get_code("1b6b002e7fb34228")
build_configs Ran Tariolle/opportunistic-target-selection/analysis/analyze_margin.py
code served (permissive licence) · get_code("17d6723584b9dc2f")
build_pairs Ran Tariolle/opportunistic-target-selection/analysis/analyze_lockmatch.py
code served (permissive licence) · get_code("ec51ead6890067f1")
compute_entropy_trajectory Ran Tariolle/opportunistic-target-selection/analysis/analyze_robust_landscape.py
code served (permissive licence) · get_code("e56096db231fc40b")
compute_perseed_tests Ran Tariolle/opportunistic-target-selection/analysis/analyze_multiseed.py
code served (permissive licence) · get_code("ecc4531cfdb06095")
compute_pooled_test Ran Tariolle/opportunistic-target-selection/analysis/analyze_multiseed.py
code served (permissive licence) · get_code("9d7f8c83b6f60a92")
fig_lockmatch_savings Ran Tariolle/opportunistic-target-selection/analysis/analyze_lockmatch.py
code served (permissive licence) · get_code("bee3bb1c9640269f")
fig_overlap Ran Tariolle/opportunistic-target-selection/analysis/analyze_target_overlap.py
code served (permissive licence) · get_code("1014490494228531")
fig_per_model Ran Tariolle/opportunistic-target-selection/analysis/analyze_benchmark.py
code served (permissive licence) · get_code("0fd074a6d3f951eb")
filter_runs Ran Tariolle/opportunistic-target-selection/analysis/analyze_robust_landscape.py
code served (permissive licence) · get_code("fee328f1f86bd7a6")
load_csv Ran Tariolle/opportunistic-target-selection/analysis/analyze_lockmatch.py
code served (permissive licence) · get_code("7e89e69f7971931c")
load_data Ran Tariolle/opportunistic-target-selection/analysis/analyze_benchmark.py
code served (permissive licence) · get_code("43e33823b698bde0")
load_data Ran Tariolle/opportunistic-target-selection/analysis/analyze_oracle_beat.py
code served (permissive licence) · get_code("6c723a46b19147df")
load_data Ran Tariolle/opportunistic-target-selection/analysis/analyze_winrate.py
code served (permissive licence) · get_code("b7714b19c17b22dd")
load_existing_keys Ran Tariolle/opportunistic-target-selection/benchmarks/ablation_target_selection.py
code served (permissive licence) · get_code("8cb86b7f243f1725")
load_json_runs Ran Tariolle/opportunistic-target-selection/analysis/analyze_robust_landscape.py
code served (permissive licence) · get_code("480b7c94873ce0ad")
load_margin_csv Ran Tariolle/opportunistic-target-selection/analysis/analyze_margin.py
code served (permissive licence) · get_code("22c29bc74df94117")
load_winrate_csv Ran Tariolle/opportunistic-target-selection/analysis/analyze_margin.py
code served (permissive licence) · get_code("c51b017253ac26a9")
summarize Ran Tariolle/opportunistic-target-selection/analysis/analyze_oracle_beat.py
code served (permissive licence) · get_code("f765883ce1ccc896")
compute_oracle_beat Not yet run Tariolle/opportunistic-target-selection/analysis/analyze_oracle_beat.py
code served (permissive licence) · get_code("0aec3be9ec99d4a6")
fig_headline_bars Not yet run Tariolle/opportunistic-target-selection/analysis/analyze_benchmark.py
code served (permissive licence) · get_code("10113eb6963af894")
select_clean_argmax Not yet run Tariolle/opportunistic-target-selection/benchmarks/ablation_target_selection.py
code served (permissive licence) · get_code("d904b3c28e4a6340")

Repositories linked to this paper

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

Abstract

Black-box adversarial attacks that minimize only the ground-truth confidence suffer from class drift: perturbations wander through the feature space without committing to a specific adversarial class, wasting queries on diffuse, undirected progress. We introduce Opportunistic Target Selection (OTS), a lightweight wrapper that switches an untargeted attack to a targeted objective early in its trajectory, locking onto whichever non-true class currently leads. OTS requires no architectural modification to the underlying attack, no gradient access, and no a priori target-class knowledge. We validate OTS on three score-based attacks (SimBA, Square Attack with cross-entropy loss, and Bandits) across five standard ImageNet classifiers (4,500 runs). On random-search attacks, OTS closely tracks oracle performance, with gains up to +27 pp in success rate and 43% relative reduction in censored-mean iterations on ResNet-50. On gradient-estimation attacks (Bandits) and attacks with margin loss, OTS is redundant, a negative result that reinforces our interpretation of OTS as a margin-loss surrogate. On adversarially-trained models, a bimodal difficulty distribution eliminates the regime where targeting helps. We release our code at https://github.com/Tariolle/opportunistic-target-selection.

For agents

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

get_harvested_code_for_paper("2605.25663")
get_code_for_paper("2605.25663")
have("2605.25663")

Connect an agent — have() is free.