Directional Reasoning Injection for Fine-Tuning MLLMs

• URL: http://arxiv.org/abs/2510.15050v1
• Date: Thu, 16 Oct 2025 18:06:46 GMT
• Title: Directional Reasoning Injection for Fine-Tuning MLLMs
• Authors: Chao Huang, Zeliang Zhang, Jiang Liu, Ximeng Sun, Jialian Wu, Xiaodong Yu, Ze Wang, Chenliang Xu, Emad Barsoum, Zicheng Liu,
• Abstract summary: Multimodal large language models (MLLMs) are rapidly advancing, yet their reasoning ability often lags behind that of strong text-only counterparts.<n>Existing methods to bridge this gap rely on supervised fine-tuning over large-scale multimodal reasoning data or reinforcement learning.<n>We propose Directional Reasoning Injection for Fine-Tuning (DRIFT) to solve this problem.
• Score: 51.53222423215055
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multimodal large language models (MLLMs) are rapidly advancing, yet their reasoning ability often lags behind that of strong text-only counterparts. Existing methods to bridge this gap rely on supervised fine-tuning over large-scale multimodal reasoning data or reinforcement learning, both of which are resource-intensive. A promising alternative is model merging, which interpolates parameters between reasoning-enhanced LLMs and multimodal variants. However, our analysis shows that naive merging is not always a "free lunch": its effectiveness varies drastically across model families, with some (e.g., LLaVA, Idefics) benefiting while others (e.g., Qwen) suffer performance degradation. To address this, we propose Directional Reasoning Injection for Fine-Tuning (DRIFT) MLLMs, a lightweight method that transfers reasoning knowledge in the gradient space, without destabilizing multimodal alignment. DRIFT precomputes a reasoning prior as the parameter-space difference between reasoning and multimodal variants, then uses it to bias gradients during multimodal fine-tuning. This approach preserves the simplicity of standard supervised fine-tuning pipelines while enabling efficient reasoning transfer. Extensive experiments on multimodal reasoning benchmarks, including MathVista and MathVerse, demonstrate that DRIFT consistently improves reasoning performance over naive merging and supervised fine-tuning, while matching or surpassing training-heavy methods at a fraction of the cost.

Related papers

• $\ abla$-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space [71.23672814629448]
  $nabla$-Reasoner is an iterative generation framework that integrates differentiable optimization over token logits into the decoding loop.<n>$nabla$-Reasoner achieves over 20% accuracy improvement on a challenging mathematical reasoning benchmark.
  arXiv  Detail & Related papers  (2026-03-05T08:42:54Z)
• Steering Large Reasoning Models towards Concise Reasoning via Flow Matching [18.79674738541318]
  FlowSteer is a nonlinear steering method that learns a complete transformation between the distributions associated with verbose and concise reasoning.<n>Our work demonstrates that modeling the full distributional transport with generative techniques offers a more effective and principled foundation for controlling LRMs.
  arXiv  Detail & Related papers  (2026-02-05T10:56:13Z)
• Latent-Space Contrastive Reinforcement Learning for Stable and Efficient LLM Reasoning [16.244366307890832]
  We propose textbfDeepLatent Reasoning (DLR), a latent-space bidirectional contrastive reinforcement learning framework.<n>This framework shifts the trial-and-error cost from expensive token-level full sequence generation to the continuous latent manifold.<n> Experiments demonstrate that DLR achieves more stable training convergence, supports longer-horizon reasoning chains, and facilitates the sustainable accumulation of reasoning capabilities.
  arXiv  Detail & Related papers  (2026-01-24T03:18:22Z)
• PA-FAS: Towards Interpretable and Generalizable Multimodal Face Anti-Spoofing via Path-Augmented Reinforcement Learning [42.24912525813944]
  Face anti-spoofing (FAS) has recently advanced in multimodal fusion, cross-domain generalization, and interpretability.<n>We propose PA-FAS, which enhances reasoning paths by constructing high-quality extended reasoning sequences from limited annotations.<n>We also introduce an answer-shuffling mechanism during SFT to force comprehensive multimodal analysis instead of using superficial cues.
  arXiv  Detail & Related papers  (2025-11-22T05:55:08Z)
• CoT Vectors: Transferring and Probing the Reasoning Mechanisms of LLMs [33.63911145333626]
  Chain-of-Thought prompting has emerged as a powerful approach to enhancing the reasoning capabilities of Large Language Models.<n>Existing implementations, such as in-context learning and fine-tuning, remain costly and inefficient.<n>We introduce CoT Vectors, compact representations that encode task-general, multi-step reasoning knowledge.
  arXiv  Detail & Related papers  (2025-10-01T06:58:23Z)
• Fractional Reasoning via Latent Steering Vectors Improves Inference Time Compute [60.151643048803145]
  We propose Fractional Reasoning, a framework that enables continuous control over reasoning intensity at inference time.<n>Our method operates by extracting the latent steering vector associated with deeper reasoning and reapplying it with a tunable scaling factor.<n> Experiments on GSM8K, MATH500, and GPQA demonstrate that Fractional Reasoning consistently improves performance across diverse reasoning tasks and models.
  arXiv  Detail & Related papers  (2025-06-18T21:15:59Z)
• Perceptual Decoupling for Scalable Multi-modal Reasoning via Reward-Optimized Captioning [78.17782197231325]
  We propose a reasoning-guided reinforcement learning strategy that aligns the extractor's captioning behavior with the reasoning objective.<n> Experiments on multi-modal math and science benchmarks show that the proposed RACRO method achieves state-of-the-art average performance.
  arXiv  Detail & Related papers  (2025-06-05T02:28:07Z)
• Prolonged Reasoning Is Not All You Need: Certainty-Based Adaptive Routing for Efficient LLM/MLLM Reasoning [27.498043430208085]
  Excessive reliance on chain-of-thought (CoT) reasoning can impair model performance.<n>We propose Certainty-based Adaptive Reasoning (CAR)<n>CAR switches between short answers and long-form reasoning based on the model perplexity.
  arXiv  Detail & Related papers  (2025-05-21T06:20:17Z)
• Taming Flow Matching with Unbalanced Optimal Transport into Fast Pansharpening [10.23957420290553]
  We propose the Optimal Transport Flow Matching framework to achieve one-step, high-quality pansharpening.<n>The OTFM framework enables simulation-free training and single-step inference while maintaining strict adherence to pansharpening constraints.
  arXiv  Detail & Related papers  (2025-03-19T08:10:49Z)
• SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMs [48.28847964704554]
  Chain-of-Thought (CoT) reasoning enables Large Language Models (LLMs) to solve complex reasoning tasks.<n>We propose a novel approach for continuous-space reasoning that does not require modifying the LLM.
  arXiv  Detail & Related papers  (2025-02-17T18:52:29Z)
• Disperse-Then-Merge: Pushing the Limits of Instruction Tuning via Alignment Tax Reduction [75.25114727856861]
  Large language models (LLMs) tend to suffer from deterioration at the latter stage ofSupervised fine-tuning process. We introduce a simple disperse-then-merge framework to address the issue. Our framework outperforms various sophisticated methods such as data curation and training regularization on a series of standard knowledge and reasoning benchmarks.
  arXiv  Detail & Related papers  (2024-05-22T08:18:19Z)
• Large Language Models as an Indirect Reasoner: Contrapositive and Contradiction for Automated Reasoning [74.90592233107712]
  We propose a Direct-Indirect Reasoning (DIR) method, which considers Direct Reasoning (DR) and Indirect Reasoning (IR) as multiple parallel reasoning paths that are merged to derive the final answer.<n>Our DIR method is simple yet effective and can be straightforwardly integrated with existing variants of CoT methods.
  arXiv  Detail & Related papers  (2024-02-06T03:41:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.