跳转至

TreeFI：面向深度神经网络的价值感知统计故障注入

文章背景与核心概要

在硬件故障下评估深度神经网络（DNN）的可靠性通常依赖于故障注入（FI）。然而，对于现代模型和数据集而言，穷举式的实验在计算上是不可行的，而传统的统计故障注入仍然需要大量的注入预算。本文介绍了一种名为 TreeFI 的新型价值感知统计故障注入方法，专门用于处理 DNN 激活值和权重中的 FP32 单比特故障。

通过利用回归树将层值分布划分为具有相似期望比特翻转行为的区间，并将注入策略性地分配给对故障率估计最具关键性的区域，TreeFI 在严格维持置信度和误差范围的同时，大幅减少了所需的注入预算。实验表明，TreeFI 将所需的注入预算最高减少了 72.1倍，激活值故障的平均减少幅度达 44.9倍，已执行权重实验的平均减少幅度达 11.2倍。该研究已被计算机辅助设计顶级会议 ICCAD 2026 接受。

摘要 (Summary)

Evaluating the reliability of deep neural networks (DNNs) under hardware faults typically relies on fault injection (FI). However, exhaustive campaigns are computationally intractable for modern models and datasets, and traditional statistical fault injection still requires large injection budgets. TreeFI is a novel, value-aware statistical fault-injection methodology designed for FP32 single-bit faults in DNN activations and weights. By leveraging regression trees to partition layer value distributions into intervals with similar expected bit-flip behaviors—and strategically allocating injections to regions most critical for failure-rate estimation—TreeFI drastically reduces the required injection budget while strictly maintaining confidence and error margins.

在硬件故障下评估深度神经网络（DNN）的可靠性通常依赖于故障注入（FI）。然而，对于现代模型和数据集而言，穷举式的实验在计算上是不可行的，而传统的统计故障注入仍然需要大量的注入预算。TreeFI 是一种新颖的、具备价值感知的统计故障注入方法，专为 DNN 激活值和权重中的 FP32 单比特故障而设计。通过利用回归树将层值分布划分为具有相似期望比特翻转行为的区间——并将注入策略性地分配给对故障率估计最具关键性的区域，TreeFI 在严格维持置信度和误差范围的同时，大幅减少了所需的注入预算。

元数据 (Metadata)

  • arXiv ID: arXiv:2609.04912 [cs.AR]
  • Authors: Noam Bires, Marcello Traiola, Angeliki Kritikakou, Elisa Fromont
  • Submitted: September 4, 2026
  • Conference Acceptance: Accepted at ICCAD 2026
  • Primary Subject: Hardware Architecture (cs.AR)
  • Secondary Subject: Artificial Intelligence (cs.AI)
  • Full-Text Resources: View PDF | TeX Source
  • arXiv ID: arXiv:2609.04912 [cs.AR]
  • 作者: Noam Bires, Marcello Traiola, Angeliki Kritikakou, Elisa Fromont
  • 提交时间: 2026年9月4日
  • 会议录用: 已被 ICCAD 2026 接受
  • 主要学科: 硬件架构 (cs.AR)
  • 次要学科: 人工智能 (cs.AI)
  • 全文资源: 查看 PDF | TeX 源码

摘要详情 (Abstract)

Reliability evaluation of deep neural networks under hardware faults commonly relies on fault injection, but exhaustive campaigns are intractable for modern models and datasets. Statistical fault injection reduces this cost, yet existing approaches still require large injection budgets because they do not explicitly exploit a key property of floating-point faults: the effect of a bit flip depends strongly on the value being corrupted.

We propose TreeFI, a value-aware statistical fault-injection methodology for FP32 single-bit faults in DNN activations and weights. TreeFI partitions each layer's value distribution into intervals with similar expected bit-flip behavior, learned using regression trees, and allocates injections across these intervals according to their relevance for failure-rate estimation. This stratified allocation preserves the target confidence and error margin while avoiding unnecessary injections in low-impact regions of the fault space.

We validate TreeFI on CNN and Transformer models using CIFAR-10 and ImageNet. On ResNet8, where exhaustive activation fault injection is feasible, TreeFI provides more accurate estimates than state-of-the-art statistical FI baselines under the same campaign setting. Across the evaluated models, TreeFI reduces the required injection budget by up to 72.1x, with average reductions of 44.9x for activation faults and 11.2x for the executed weight campaigns.

在硬件故障下对深度神经网络进行可靠性评估通常依赖于故障注入，但对于现代模型和数据集来说，穷举式实验在计算上是无法实现的。统计故障注入降低了这一成本，但现有方法仍然需要大量的注入预算，因为它们没有明确利用浮点数故障的一个关键特性：比特翻转的影响在很大程度上取决于被破坏的值。

我们提出了 TreeFI，这是一种针对 DNN 激活值和权重中 FP32 单比特故障的价值感知统计故障注入方法。TreeFI 利用回归树将每层的数值分布划分为具有相似期望比特翻转行为的区间，并根据这些区间对故障率估计的相关性分配注入。这种分层分配在保持目标置信度和误差范围的同时，避免了在故障空间中低影响区域进行不必要的注入。

我们在使用 CIFAR-10 和 ImageNet 的 CNN 和 Transformer 模型上验证了 TreeFI。在穷举激活值故障注入可行的 ResNet8 上，在相同的实验设置下，TreeFI 提供了比现有最先进的统计 FI 基线更准确的估计。在所评估的模型中，TreeFI 将所需的注入预算减少了高达 72.1 倍，其中激活值故障平均减少 44.9 倍，已执行的权重实验平均减少 11.2 倍。

(License: Creative Commons Attribution 4.0 International)