REVIEW 10 cited by
Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start
read the original abstract
Recent advancements in large language models (LLMs) have demonstrated impressive chain-of-thought reasoning capabilities, with reinforcement learning (RL) playing a crucial role in this progress. While "aha moment" patterns--where models exhibit self-correction through reflection--are often attributed to emergent properties from RL, we first demonstrate that these patterns exist in multimodal LLMs (MLLMs) prior to RL training but may not necessarily correlate with improved reasoning performance. Building on these insights, we present a comprehensive study on enhancing multimodal reasoning through a two-stage approach: (1) supervised fine-tuning (SFT) as a cold start with structured chain-of-thought reasoning patterns, followed by (2) reinforcement learning via GRPO to further refine these capabilities. Our extensive experiments show that this combined approach consistently outperforms both SFT-only and RL-only methods across challenging multimodal reasoning benchmarks. The resulting models achieve state-of-the-art performance among open-source MLLMs at both 3B and 7B scales, with our 7B model showing substantial improvements over base models (e.g., 66.3 %$\rightarrow$73.4 % on MathVista, 62.9 %$\rightarrow$70.4 % on We-Math) and our 3B model achieving performance competitive with several 7B models. Overall, this work provides practical guidance for building advanced multimodal reasoning models. Our code is available at https://github.com/waltonfuture/RL-with-Cold-Start.
Forward citations
Cited by 10 Pith papers
-
ProtoCycle: Reflective Tool-Augmented Planning for Text-Guided Protein Design
ProtoCycle improves text-guided protein design by coupling an LLM planner with tool feedback and reflection to achieve better language alignment and foldability than direct generation.
-
Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation
Vision-OPD transfers an MLLM's privileged regional perception to its full-image policy through on-policy token-level self-distillation, yielding competitive results on fine-grained visual benchmarks.
-
Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation
Vision-OPD uses on-policy self-distillation from crop-conditioned to full-image policies within the same MLLM to close the regional-to-global perception gap.
-
Self-Consistent Latent Reasoning: Long Latent Sequence Reasoning for Vision-Language Model
SCOLAR addresses information gain collapse in latent visual reasoning by generating independent auxiliary visual tokens from LLM hidden states, extending acceptable CoT length over 30x and achieving +14.12% gains on b...
-
Self-Consistent Latent Reasoning: Long Latent Sequence Reasoning for Vision-Language Model
SCOLAR fixes information gain collapse in latent visual reasoning by generating independent auxiliary visual tokens via a detransformer, extending acceptable CoT length over 30x and delivering +14.12% gains on reasoni...
-
LaRe: Latent Refocusing for Multimodal Reasoning
LaRe performs iterative visual refocusing in latent space and reports accuracy gains with fewer tokens, but its main experiments compare against baselines trained with less data.
-
SLMs as Multi-Agent Routers: A Progressive SFT and Reinforcement Learning Approach
A 0.6B router trained by SFT+RL on retrieval-quality rewards reaches 0.771 NDCG@10 across 11 agents, beating intent-prompted LLMs and cutting latency by 82%.
-
Be Faithful When Response: Returning Fluent and Grounded Answers for Vision-Language Models Reinforcement Learning
Faithful Warm-Start pre-training on causally consistent vision-language samples improves accuracy, stabilizes RL, and reduces unsupported reasoning in VLMs.
-
V-Zero: Answer-Label-Free On-Policy Distillation with Contrastive Evidence Gating for Fine-Grained Visual Reasoning
V-Zero trains MLLMs for visual reasoning without answer labels by gating on-policy distillation trajectories using contrastive evidence from relevant versus negative image crops.
-
Distilling Game Code World Model Generation into Lightweight Large Language Models
SFT followed by RLVR on Qwen2.5-3B-Instruct raises syntactic and execution correctness when generating Game Code World Models across 30 games.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.