Built on

“Visualizing Attention in Transformer-Based Language Representation Models” arXiv:1904.02679 [cs]
Original
Jesse Vig · 1904
Earlier work this paper cites.
“Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI” arXiv:1910.10045 [cs]
Original
Alejandro Arrieta et al · 1910
Earlier work this paper cites.
“exBERT: A Visual Analysis Tool to Explore Learned Representations in Transformers Models” arXiv:1910.05276 [cs]
Original
Benjamin Hoover, Hendrik Strobelt and Sebastian Gehrmann · 1910
Earlier work this paper cites.
“A Unified Approach to Interpreting Model Predictions” arXiv:1705.07874 [cs]
Original
Scott Lundberg and Su-In Lee · 2017
Earlier work this paper cites.
“Understanding intermediate layers using linear classifier probes” arXiv:1610.01644 [stat]
Original
Guillaume Alain and Yoshua Bengio · 2018
Earlier work this paper cites.
“Evolutionary Fuzzy Systems for Explainable Artificial Intelligence: Why, When, What for, and Where to?”
Alberto Fernandez et al · 2018
Earlier work this paper cites.
“Deep k-Nearest Neighbors: Towards Confident, Interpretable and Robust Deep Learning” arXiv:1803.04765 [cs]
Original
Nicolas Papernot and Patrick McDaniel · 2018
Earlier work this paper cites.
“Seq2Seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models” arXiv:1804.09299 [cs]
Original
Hendrik Strobelt et al · 2018
Earlier work this paper cites.
“Extending Knowledge Graphs with Subjective Influence Networks for Personalized Fashion”
Kurt Bollacker, Natalia Díaz-Rodríguez and Xian Li · 2019
Earlier work this paper cites.
“Model Explainability in Deep Learning Based Natural Language Processing” arXiv:2106.07410 [cs]
Original
Shafie Gholizadeh and Nengfeng Zhou · 2021
Earlier work this paper cites.
“Interactive Visualization and Manipulation of Attention-based Neural Machine Translation”
Jaesong Lee, Joong-Hwi Shin and Jun-Seok Kim · 2021
Earlier work this paper cites.

Similar

“Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small” arXiv:2211.00593 [cs]
Original
Kevin Wang et al · 2022
Cited alongside, same era.
“Prompt-Augmented Linear Probing: Scaling beyond the Limit of Few-shot In-Context Learners” arXiv:2212.10873 [cs]
Original
Hyunsoo Cho et al · 2023
Cited alongside, same era.
“Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations” arXiv:2312.06674 [cs]
Original
Hakan Inan et al · 2023
Cited alongside, same era.
“Prompt Packer: Deceiving LLMs through Compositional Instruction with Hidden Attacks” arXiv:2310.10077 [cs]
Original
Shuyu Jiang, Xingshu Chen and Rui Tang · 2023
Cited alongside, same era.
“LLM-Pruner: On the Structural Pruning of Large Language Models” arXiv:2305.11627 [cs]

Then

“The Llama 3 Herd of Models” arXiv:2407.21783 [cs]
Original
Aaron Grattafiori et al · 2024
Closest in time.
“The Unreasonable Ineffectiveness of the Deeper Layers” arXiv:2403.17887 [cs]
Original
Andrey Gromov et al · 2024
Closest in time.
“Attention Tracker: Detecting Prompt Injection Attacks in LLMs” arXiv:2411.00348 [cs]
Original
Kuo-Han Hung et al · 2024
Closest in time.
“Large Language Models Are Overparameterized Text Encoders” arXiv:2410.14578 [cs]
Original
Thennal. K, Tim Fischer and Chris Biemann · 2024
Closest in time.
Lakera, 2024
“lakeraai/pint-benchmark” original-date: 2024-03-27T19:04:05Z · 2024
Closest in time.

Beyond the bibliography

alphaXiv searches the wider corpus for related work and actual follow-ups.

Open on alphaXiv

alphaXiv is searching for related work…

Reading the bibliography…

2024

Lightweight Safety Classification Using Pruned Language Models

Sawtell, Mason, Masterman, Tula, Besen, Sandi et al.

Understand

In this paper, we introduce a novel technique for content safety and prompt injection classification for Large Language Models.

Our technique, Layer Enhanced Classification (LEC), trains a Penalized Logistic Regression (PLR) classifier on the hidden state of an LLM's optimal intermediate transformer layer.
By combining the computational efficiency of a streamlined PLR classifier with the sophisticated language understanding of an LLM, our approach delivers superior performance surpassing GPT-4o and special-purpose models fine-tuned for each task.
We find that small general-purpose models (Qwen 2.5 sizes 0.5B, 1.5B, and 3B) and other transformer-based architectures like DeBERTa v3 are robust feature extractors allowing simple classifiers to be effectively trained on fewer than 100 high-quality examples.