AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Language Models are Injective and Hence Invertible

Giorgos Nikolaou, Tommaso Mencattini, Donato Crisostomi, Andrea Santilli, Yannis Panagakis, Emanuele Rodolà

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Efficient language transformers · 40 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
use FP4 weight quantization instead of FP32Llama-3.1-8B · Collision search, final layermin pairwise l2 distance: 1.274 -> 2.281Improved
use INT8 weight quantization instead of FP32Llama-3.1-8B · Collision search, final layermin pairwise l2 distance: 1.274 -> 6.597Improved
use FP4 weight quantization instead of FP32Mistral-7B-v0.1 · Collision search, final layermin pairwise l2 distance: 1.136 -> 1.748Improved
use INT8 weight quantization instead of FP32Phi-4-mini-instruct · Collision search, final layermin pairwise l2 distance: 8.780 -> 20.956Improved
go from first layer to last layerLlama-3.1-8B · Collision searchmin pairwise l2 distance: 0.001 -> 0.620Improved
go from first layer to last layerLlama-3.1-70B · Collision searchmin pairwise l2 distance: 0.005 -> 3.975Improved
use SipIt instead of BruteForceprompt inversion · Prompt inversion, 20 tokensmean time (s): 3889.61 -> 28.01Improved
use SipIt instead of HardPromptsprompt inversion · Prompt inversion, 20 tokensaccuracy: 0.00 -> 1.00Improved
increase vocabulary from 32000 to 128255SipIt with FP4 models · SipIt inversion, FP4vocab explored (%): 0.19 ± 0.08 % -> 0.21 ± 0.10 %No clear effect
increase prompt length (GPT-2 Small)last-token states · Collision searchmin pairwise l2 distance: rises at short lengths then stabilizesMixed
deeper layer for inversionSipIt · Inversion time, prompts 20-200 tokenstime (s): runtimes rise only slightlyNo 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