On the Robustness of CountSketch to Adaptive Inputs

Authors: Edith Cohen, Xin Lyu, Jelani Nelson, Tamas Sarlos, Moshe Shechner, Uri Stemmer

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We complement our analysis with empirical evaluation, showing that our attack is practical. ... Code for all the experiments is available at https: //github.com/google-research/google-research/ tree/master/robust_count_sketch.
Researcher Affiliation Collaboration 1Google Research 2Tel Aviv University 3UC Berkeley.
Pseudocode Yes Algorithm 1: Threshold Monitor (34) ... Algorithm 2: Robust Threshold BCount Sketch Estimator ... Algorithm 3: Weight Estimator ... Algorithm 4: Robust BCount Sketch: Streaming
Open Source Code Yes Code for all the experiments is available at https: //github.com/google-research/google-research/ tree/master/robust_count_sketch.
Open Datasets No The paper describes generating input vectors for simulation ('input vectors with one 'heavy' entry and n i.i.d. entries N(0, 1)') but does not explicitly state the use of a publicly available or open dataset, nor does it provide a link or citation for one.
Dataset Splits No The paper does not provide explicit training/test/validation dataset splits for its own experiments. It only mentions 'simulation results' and describes how input vectors were generated.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments.
Software Dependencies No The paper does not provide specific software names with version numbers for its dependencies.
Experiment Setup Yes We performed additional simulations (not shown) where we swept b from 30 to 300, keeping d/b = 100 and keeping k /b = 3. ... The evaluation used input vectors with one 'heavy' entry and n i.i.d. entries N(0, 1). ... Sketch parameters b = 7, d/b = 101, left: n = 5 104, right: n = 1 105. (Figure 2 caption) ... We use b = 7. Left: v = 20, n = 5 103 middle: v = 25, n = 1 104 right: v = 70, n = 5 104. (Figure 3 caption)