---
license: mit
language:
- en
base_model:
- FacebookAI/roberta-base
pipeline_tag: feature-extraction
---

Consider this to be an alpha version of the LUSR model. 

This checkpoint outperforms LUAR in machine-text detection and a task probes for stylistic features (mSTEL). 

### Contrastive Probing for Stylistic Features

| Model Name                     | Accuracy |
|--------------------------------|----------|
| LUAR-MUD                       | 3        |
| AnnaWegmann/Style-Embedding    | 22.4     |
| LUSR                           | 22.2     |


### Few-Shot Machine-text Detection, Query Size = 1

| Model Name     | pAUC(0.01) N=1 | N=2  | N=5  |
|----------------|----------------|------|------|
| Support Size   | N=1            | N=2  | N=5  |
| LUAR-MUD       | 0.63           | 0.74 | 0.89 |
| Anna Wegmann   | 0.58           | 0.69 | 0.84 |
| LUSR           | **0.70**       | **0.83** | **0.97** |