johannes-garstenauer/embeddings_from_distilbert_masking_heaps

--- dataset_info: features: - name: struct dtype: string - name: label dtype: int64 - name: pred dtype: int64 - name: cls_layer_6 sequence: float32 - name: cls_layer_5 sequence: float32 - name: cls_layer_4 sequence: float32 splits: - name: train num_bytes: 2564388529 num_examples: 269087 download_size: 2984971174 dataset_size: 2564388529 --- # Dataset Card for "embeddings_from_distilbert_masking_heaps" Dataset created for thesis: "Generating Robust Representations of Structures in OpenSSH Heap Dumps" by Johannes Garstenauer. This dataset contains representations of heap data structures along with their labels and the predicted label. The representations are the [CLS] token embeddings of the last 3 layers of the DistilBERT model. The representation-generating model is: https://huggingface.co/johannes-garstenauer/distilbert_masking_heaps The dataset from which representations were generated is: https://huggingface.co/datasets/johannes-garstenauer/structs_token_size_4_reduced_labelled_eval The twin dataset (model with different training used to generate embeddings) is: https://huggingface.co/datasets/johannes-garstenauer/embeddings_from_distilbert_class_heaps/ Thesis and associated scripts: https://zenodo.org/records/10053730

原始信息汇总

数据集概述

数据集名称

• 名称: embeddings_from_distilbert_masking_heaps

数据集创建目的

• 目的: 用于论文 "Generating Robust Representations of Structures in OpenSSH Heap Dumps" 中，作者为 Johannes Garstenauer。

数据集内容

• 内容: 包含堆数据结构的表示、标签及预测标签。这些表示是 DistilBERT 模型最后三层的 [CLS] 令牌嵌入。

数据集特征

• 特征列表:
  • struct: 数据类型为 string
  • label: 数据类型为 int64
  • pred: 数据类型为 int64
  • cls_layer_6: 数据类型为 float32 的序列
  • cls_layer_5: 数据类型为 float32 的序列
  • cls_layer_4: 数据类型为 float32 的序列

数据集分割

• 训练集:
  • 名称: train
  • 字节数: 2564388529
  • 样本数: 269087

数据集大小

• 下载大小: 2984971174
• 数据集大小: 2564388529