AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

An Invariant Latent Space Perspective on Language Model Inversion

Wentao Ye, Jiaqi Hu, Haobo Wang, Xinpeng Ti, Zhiqing Xiao, Hao Chen, Liyao Li, Lei Feng, Sai Wu, Junbo Zhao

AAAI 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Probabilistic embeddings and relational inference · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add inverse encoder in front of the frozen LLM decoder (ablation)Inv2A · User prompt inversionBLEU: 1.31 -> 35.20Improved
add inverse encoder instead of LoRA-tuned decoder alone (ablation)Inv2A · User prompt inversionBLEU: 26.47 -> 35.20Improved
use the raw forward LLM as decoder instead of Qwen2 (ablation)Inv2A · User prompt inversionBLEU: 33.38 -> 35.20Improved
add source-aware contrastive learning in alignment phase (ablation)Inv2A · User prompt inversionBLEU: 33.91 -> 35.20Improved
add optional dynamic filter post-refinement (1 round) (ablation)Inv2A · User prompt inversionBLEU: 35.20 -> 35.97Improved
add a second refinement round (ablation)Inv2A with refine · User prompt inversionBLEU: 35.97 -> 36.06No clear effect
naive round-trip decoding of outputs through the LLMraw LLM as inverse · Alpaca (GPT-2 analysis)BLEU: BLEU 4.75Hurt
raise sampling temperature as a defense against inversionInv2A vs Output2prompt · User prompt inversion, temperature 0.5 to 2.0BLEU: 35.42 -> 30.10Mixed
inject Gaussian noise into MLP layer of first hidden layer as defenseLLM defense · User prompt inversionBLEU: ~8% BLEU drop at 2.5e-2Mixed
reuse LLM latent space with Inv2A instead of training external T5 inverse modelinversion attack · training data scaleBLEU: comparable performance with 20 to 30% of the dataImproved
swap Inv2A for Output2prompt on out-of-domain Anthropic HH with LLaMA2-13BInv2A · Anthropic HH (LLaMA2-13B)BLEU: 29.61 -> 30.41No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy