finding

active

finding:one-stage-cot-qcm-ra-shows-12-31-accuracy-drop-vs-no-cot-qcm-a-on-scienceqa-two-stage-framework-rationale-generation-answer-inference-achieves-85-31-accuracy-with-vision-features

One-stage CoT (QCM→RA) shows 12.31% accuracy drop vs. no-CoT (QCM→A) on ScienceQA; two-stage framework (rationale generation + answer inference) achieves 85.31% accuracy with vision features

Empirical evidence that naive one-stage CoT fails in language-only setting; two-stage + vision achieves state-of-the-art.

Source paper

extracted_from

Multimodal Chain-of-Thought Reasoning in Language Models

(2023) · Zhuosheng Zhang · Aston Zhang · Mu Li · Hai Zhao +2

Neighborhood — ranked by edge-count

Communities (2)

community

Chain-of-Thought reasoning robustness & safety
members_of
CoT effects on generalization, multimodal QA accuracy, and AI safety alignment training.
Multimodal Chain-of-Thought Reasoning
members_of
Two-stage rationale-then-answer framework evaluated on ScienceQA benchmark, ~738M parameters.

Frameworks (1)

framework

two-stage separation of rationale generation and answer inference
supports
Architectural design principle that decouples rationale generation (stage 1) from answer inference (stage 2) in Multimodal-CoT.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Multimodal-CoT with vision features achieves higher validation accuracy at early training epochs (epoch 1-3) compared to one-stage and two-stage language-only baselines on ScienceQAfinding0.820
Evidence that multimodal information accelerates convergence speed during training.
VS-CoT achieves the highest Top@1 accuracy (0.348) and Pass@N accuracy (0.485) on SimpleQA among all methods testedfinding0.789
Shows VS not only maintains but can slightly improve factual accuracy compared to baseline methods
Multimodal-CoT trained with InstructBLIP/ChatGPT-generated rationales achieves 87.76% accuracy on ScienceQA, comparable to human-annotated rationale performance of 90.45%finding0.788
Evidence that Multimodal-CoT can operate without human-annotated reasoning chains by using large models to generate pseudo-rationales.
90.45% accuracy on ScienceQA benchmark with Multimodal-CoT Large (738M parameters)finding0.785
State-of-the-art result on ScienceQA; represents +3.91% improvement over prior best published result of 86.54%.
60.7% of hallucination mistakes corrected by adding vision features in two-stage framework on ScienceQAfinding0.778
Quantitative evidence that vision information mitigates hallucinated rationales; 56% of error cases contained hallucinations, 60.7% of which were resolved with vision features.
In the absence of prior preferences, Active Inference null model and Bayesian RL maintain exploration with average scores of 44.00 and 39.94 respectively, whereas Q-learning does not explore.finding0.770
Table 2 first row; reward shaping section.
Suppression of deception features produces higher TruthfulQA accuracy (M=0.44) than amplification (M=0.20), t(816)=6.76, p=1.5×10⁻¹⁰ across 29 categoriesfinding0.770
Out-of-domain generalization showing deception features track general representational honesty
Under spatio permutation controls, IIT consciousness estimates outperform Span Representation in mean AUC in several cases (LLaMA3.1-70B on Hinting and Irony, Mistral-7B on Irony, LLaMA3.1-8B on Strange Stories).finding0.766
Contrasts with temporal permutation where Span Representation dominates; suggests spatio permutation reveals different dynamics.