← Back to all news
arXiv Computation and Language By Donghao Zeng (Mila, McGill University), Nishant Subramani (Carnegie Mellon University), Tamar Rott Shaham (MIT CSAIL), Benno Krojer (Mila, McGill University, Northeastern University)

Cross-modal Steering Vectors Exist Across Both Image and Text Generation

Read the original on arXiv Computation and Language →

The paper investigates whether steering vectors—specific vectors that influence generation—can transfer across modalities in unified multimodal models. By demonstrating that vectors that steer text generation can also steer image generation (and vice versa) for concepts like emotion and object size, the authors show that these vectors have a consistent, interpretable effect even for modality‑specific concepts. The study reveals a geometric signature: highly transferable steering vectors point in similar directions across the modality gap, suggesting shared underlying mechanisms.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computation and Language.

Related stories

arXiv Computation and Language
Oct 12

TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs

TokenSwap introduces a method to benchmark and reduce the modality gap in multimodal large language models (MLLMs) by interleaving semantically aligned images with text tokens. By transforming text-only benchmarks like MMLU into image‑interleaved versions (TokenSwap‑Bench), the authors reveal a pervasive performance drop—averaging 19.6%—across 42 MLLMs, with reasoning models showing smaller gaps. Training with TokenSwap mitigates this gap while maintaining strong text‑only and vision‑language performance.

By Andong Hua, Colton Bishop, Igor Mordatch, Arian Hosseini, Jindong Gu, Aleksandra Faust, Rebecca Roelofs, Yao Qin
llmsmultimodalbenchmarkssafety
More like this →
Hugging Face Trending Papers
Sep 8

Studying Image Tokenizers as Visual Languages in Unified Multimodal Models

The paper investigates image tokenizers as the visual language of unified multimodal models by creating a controlled autoregressive testbed that tracks task‑specific validation losses during multimodal continual pretraining across text, image, text‑to‑image, and image‑to‑text predictions. It shows that losses must be analyzed by task, that the loss–performance relationship varies with the token space, and that better reconstruction does not always lead to stronger downstream performance. The study also demonstrates how tokenizer design choices—such as discriminator use, semantic supervision, and vocabulary size—affect joint modeling and downstream results.

diffusionnlpfine-tuningmultimodal
More like this →
arXiv AI
Oct 16

MSP: Modality Self-Play for Training Multimodal LLMs Without Paired Cross-Modal Data

The paper introduces Modality Self-Play (MSP), a training framework that enables multimodal large language models to learn cross‑modal interactions without requiring paired non‑text data. MSP uses proxy‑mediated textual descriptions of unseen modalities as semantic bridges, allowing a model to generate and align these proxies with observed modalities. Experiments show that MSP achieves state‑of‑the‑art or competitive performance on several multimodal benchmarks and demonstrates effective zero‑shot modality composition through learned bridges and anchor‑supporting proxy descriptions.

By Robert Moseley, Md Kaykobad Reza, Ameya Patil, Edward Ayrapetian, Salman Asif
llmsreinforcement-learningmultimodalbenchmarkssafety
More like this →