This study evaluates a transformer-based framework for detecting anomalies in large-scale system logs. Experiments were conducted on four public datasets—HDFS, BGL, Spirit, and Thunderbird—using adaptive log-sequence generation to handle varying sequence lengths and data rates. The model architecture includes two transformer encoder layers with multi-head attention and was optimized using AdamW and OneCycleLR. Implemented in PyTorch and trained on an HPC system, the setup demonstrates an efficient and scalable approach for benchmarking log anomaly detection methods.This study evaluates a transformer-based framework for detecting anomalies in large-scale system logs. Experiments were conducted on four public datasets—HDFS, BGL, Spirit, and Thunderbird—using adaptive log-sequence generation to handle varying sequence lengths and data rates. The model architecture includes two transformer encoder layers with multi-head attention and was optimized using AdamW and OneCycleLR. Implemented in PyTorch and trained on an HPC system, the setup demonstrates an efficient and scalable approach for benchmarking log anomaly detection methods.

How Transformer Models Detect Anomalies in System Logs

2025/11/04 01:52

Abstract

1 Introduction

2 Background and Related Work

2.1 Different Formulations of the Log-based Anomaly Detection Task

2.2 Supervised v.s. Unsupervised

2.3 Information within Log Data

2.4 Fix-Window Grouping

2.5 Related Works

3 A Configurable Transformer-based Anomaly Detection Approach

3.1 Problem Formulation

3.2 Log Parsing and Log Embedding

3.3 Positional & Temporal Encoding

3.4 Model Structure

3.5 Supervised Binary Classification

4 Experimental Setup

4.1 Datasets

4.2 Evaluation Metrics

4.3 Generating Log Sequences of Varying Lengths

4.4 Implementation Details and Experimental Environment

5 Experimental Results

5.1 RQ1: How does our proposed anomaly detection model perform compared to the baselines?

5.2 RQ2: How much does the sequential and temporal information within log sequences affect anomaly detection?

5.3 RQ3: How much do the different types of information individually contribute to anomaly detection?

6 Discussion

7 Threats to validity

8 Conclusions and References

\

4 Experimental Setup

4.1 Datasets We evaluate our proposed approach and conduct experiments with four commonlyused public datasets: HDFS [8], Blue Gene/L (BGL), Spirit, and Thunderbird [32]. These datasets are commonly used in existing studies [1, 5, 12]. The HDFS dataset [8] is derived from the Amazon EC2 platform. The dataset comprises over 11 million log events, each linked to a block ID. This block ID allows us to partition the log data into sessions. The annotations are block-wise: each session is labeled as either normal or abnormal. In total, there are 575,061 log sessions, with 16,838 (2.9%) identified as anomalies. The BGL, Spirit, and Thunderbird datasets are recorded from supercomputer systems, from which they are named. Different from the HDFS dataset, all these datasets have log item-wise annotation. However, there is no block ID or other identifier to group the log items into sequences. The BGL dataset is recorded with a time span of 215 days, containing 4,747,963 log items, where 348,460 (7.3%) are labeled as anomalies. As the Spirit and Thunderbird datasets each contain more than 200 million log items, which is too large to process, we use subsets of 5 million and 10 million log items, respectively, as per the practices of previous works [7, 11, 15]. We split the datasets into an 80% training set and a 20% test set. For the HDFS dataset, we randomly shuffle the sessions to perform dataset splitting. For the remaining datasets, we divide them in accordance with the chronological order of logs. The summarised properties of datasets utilized in the evaluation and experiment of our study are presented in Table 2.

\

4.3 Generating Log Sequences of Varying Lengths

Except for the HDFS dataset, which has a block ID to group the logs into sequences, other datasets employed by our study have no identifier to group or split the whole log sequence into sub-sequences. In practice, the logs produced by systems and applications do not adhere to a fixed rate of generation. Using fixed-window or fixed-time grouping with a sliding window fails to adequately accommodate the variability in log generation and thus may lead to inaccurate detection of anomalies in real scenarios. Moreover, according to previous studies [1, 7, 15], the best grouping setting varies depending on the dataset, and these settings can significantly influence the performance of the anomaly detection model, making it challenging to compare the effectiveness of different anomaly detection methods. Therefore, we use a method to generate log sequences with varying lengths and utilize these sequences to train the model within our anomaly detection framework. In the process of log sequence generation, we determined specific parameters, including minimum and maximum sequence lengths, as well as a designated step size. The step size is used to control the interval of the first log events in log sequences. The length of each log sequence is randomly generated in the range of the minimum and the maximum length. We assume the log sequence of the minimum length can offer a minimum context for a possible anomaly. The step size controls the overlaps of sequences. The maximum length affects the number of parameters in the model, and step size decides the amount of samples in the dataset. They should be aligned with the data distribution and computational resources available. In the experiments conducted in this study, we set the minimum length as 128, the maximum length as 512, and the step size as 64 for the datasets without a grouping identifier.

\ 4.4 Implementation Details and Experimental Environment

In our experiments, the proposed transformer-based anomaly detection model has two layers of the transformer encoder. The number of attention heads is 12, and the dimension of the feedforward network layer within each transformer block is set to 2048. We use AdamW with an initial learning rate of 5e-4 as the optimization algorithm and employ the OneCycleLR learning rate scheduler to enable a better convergence. We selected these hyperparameters following standard practices while also considering computational efficiency. Our implementation is based on Python 3.11 and PyTorch 2.2.1. All the experiments are run on a high-performance computing (HPC) system. We use a computational node equipped with an Intel Gold 6148 Skylake @ 2.4 GHz CPU, 16GB RAM and an NVIDIA V100 GPU to run our experiments.

:::info Authors:

  1. Xingfang Wu
  2. Heng Li
  3. Foutse Khomh

:::

:::info This paper is available on arxiv under CC by 4.0 Deed (Attribution 4.0 International) license.

:::

\

Disclaimer: The articles reposted on this site are sourced from public platforms and are provided for informational purposes only. They do not necessarily reflect the views of MEXC. All rights remain with the original authors. If you believe any content infringes on third-party rights, please contact service@support.mexc.com for removal. MEXC makes no guarantees regarding the accuracy, completeness, or timeliness of the content and is not responsible for any actions taken based on the information provided. The content does not constitute financial, legal, or other professional advice, nor should it be considered a recommendation or endorsement by MEXC.

You May Also Like

Tokenized Assets Shift From Wrappers to Building Blocks in DeFi

Tokenized Assets Shift From Wrappers to Building Blocks in DeFi

The post Tokenized Assets Shift From Wrappers to Building Blocks in DeFi appeared on BitcoinEthereumNews.com. RWAs are rapidly moving on-chain, unlocking new opportunities for investors and DeFi protocols, according to a new report from Dune and RWAxyz. Tokenized real-world assets (RWAs) are moving beyond digital versions of traditional securities to become key building blocks of decentralized finance (DeFi), according to the 2025 RWA Report from Dune and RWAxyz. The report notes that Treasuries, bonds, credit, and equities are now being used in DeFi as collateral, trading instruments, and yield products. This marks tokenization’s “real breakthrough” – composability, or the ability to combine and reuse assets across different protocols. Projects are already showing how this works in practice. Asset manager Maple Finance’s syrupUSDC, for example, has grown to $2.5 billion, with more than 30% placed in DeFi apps like Spark ($570 million). Centrifuge’s new deJAAA token, a wrapper for Janus Henderson’s AAA CLO fund, is already trading on Aerodrome, Coinbase and other exchanges, with Stellar planned next. Meanwhile, Aave’s Horizon RWA Market now lets institutional users post tokenized Treasuries and CLOs as collateral. This trend underscores a bigger shift: RWAs are no longer just copies of traditional assets; instead, they are becoming core parts of on-chain finance, powering lending, liquidity, and yield, and helping to close the gap between traditional finance (TradFi) and DeFi. “RWAs have crossed the chasm from experimentation to execution,” Sid Powell, CEO of Maple Finance, says in the report. “Our growth to $3.5B AUM reflects a broader shift: traditional financial services are adopting crypto assets while institutions seek exposure to on-chain markets.” Investor demand for higher returns and more diversified options is mainly driving this growth. Tokenized Treasuries proved there is strong demand, with $7.3 billion issued by September 2025 – up 85% year-to-date. The growth was led by BlackRock, WisdomTree, Ondo, and Centrifuge’s JTRSY (Janus Henderson Anemoy Treasury Fund). Spark’s $1…
Share
BitcoinEthereumNews2025/09/18 06:10