# Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs?

arXiv: https://arxiv.org/abs/2501.02669 · HTML: https://mlsanity.com/paper/2501.02669/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 20 of its 95 references that are in the map.
- First posted on arXiv 5 Jan 2025.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)

## Builds on (most cited first, 20 of 20)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [Microsoft COCO Captions: Data Collection and Evaluation Server](https://mlsanity.com/paper/1504.00325/) (2015)
- [PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel](https://mlsanity.com/paper/2304.11277/) (2023)
- [MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?](https://mlsanity.com/paper/2403.14624/) (2024)
- [Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision](https://mlsanity.com/paper/2312.09390/) (2023)
- [From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step](https://mlsanity.com/paper/2405.14838/) (2024)
- [Pixtral 12B](https://mlsanity.com/paper/2410.07073/) (2024)
- [How to Train Long-Context Language Models (Effectively)](https://mlsanity.com/paper/2410.02660/) (2024)
- [Transformers Can Achieve Length Generalization But Not Robustly](https://mlsanity.com/paper/2402.09371/) (2024)
- [Length Generalization in Arithmetic Transformers](https://mlsanity.com/paper/2306.15400/) (2023)
- [VSP: Assessing the dual challenges of perception and reasoning in spatial planning tasks for VLMs](https://mlsanity.com/paper/2407.01863/) (2024)
- [Are Language Models Puzzle Prodigies? Algorithmic Puzzles Unveil Serious Challenges in Multimodal Reasoning](https://mlsanity.com/paper/2403.03864/) (2024)
- [Cross-Modal Consistency in Multimodal Large Language Models](https://mlsanity.com/paper/2411.09273/) (2024)
- [On Pre-training of Multimodal Language Models Customized for Chart Understanding](https://mlsanity.com/paper/2407.14506/) (2024)
- [Benchmarking ChatGPT on Algorithmic Reasoning](https://mlsanity.com/paper/2404.03441/) (2024)
- [How language models extrapolate outside the training data: A case study in Textualized Gridworld](https://mlsanity.com/paper/2406.15275/) (2024)
- [Are Large-Language Models Graph Algorithmic Reasoners?](https://mlsanity.com/paper/2410.22597/) (2024)
- [Lost in Translation: When GPT-4V(ision) Can't See Eye to Eye with Text. A Vision-Language-Consistency Analysis of VLLMs and Beyond](https://mlsanity.com/paper/2310.12520/) (2023)

## Built on by (most cited first, 14 of 14)

- [LEGO-Puzzles: How Good Are MLLMs at Multi-Step Spatial Reasoning?](https://mlsanity.com/paper/2503.19990/) (2025)
- [A Frustratingly Simple Yet Highly Effective Attack Baseline: Over 90% Success Rate Against the Strong Black-box Models of GPT-4.5/4o/o1](https://mlsanity.com/paper/2503.10635/) (2025)
- [Why Reasoning Matters? A Survey of Advancements in Multimodal Reasoning (v1)](https://mlsanity.com/paper/2504.03151/) (2025)
- [OPD-V: Visual On-Policy Self-Distillation with Modality Balance](https://mlsanity.com/paper/2608.05131/) (2026)
- [SEAM: Semantically Equivalent Across Modalities Benchmark for Vision-Language Models](https://mlsanity.com/paper/2508.18179/) (2025)
- [GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models](https://mlsanity.com/paper/2605.29398/) (2026)
- [Textual-to-Visual Iterative Self-Verification for Slide Generation](https://mlsanity.com/paper/2502.15412/) (2025)
- [GamiBench: Evaluating Spatial Reasoning and 2D-to-3D Planning Capabilities of MLLMs with Origami Folding Tasks](https://mlsanity.com/paper/2512.22207/) (2025)
- [Mind the Gap Between Spatial Reasoning and Acting! Step-by-Step Evaluation of Agents With Spatial-Gym](https://mlsanity.com/paper/2604.09338/) (2026)
- [Beyond the Linear Separability Ceiling: Aligning Representations in VLMs](https://mlsanity.com/paper/2507.07574/) (2025)
- [STELAR-VISION: Self-Topology-Aware Efficient Learning for Aligned Reasoning in Vision](https://mlsanity.com/paper/2508.08688/) (2025)
- [When Visuals Aren't the Problem: Evaluating Vision-Language Models on Misleading Data Visualizations](https://mlsanity.com/paper/2603.22368/) (2026)
- [Not Another Text Benchmark: Putting the "Visual" Back in Visual Question Answering for Large Video Models](https://mlsanity.com/paper/2609.17112/) (2026)
- [Learning How to Remember: Memory Modality Selection for GUI Agents](https://mlsanity.com/paper/2610.17928/) (2026)

## Abstract (arXiv)

Vision Language Models (VLMs) are impressive at visual question answering and image captioning. But they underperform on multi-step visual reasoning -- even compared to LLMs on the same tasks presented in text form -- giving rise to perceptions of modality imbalance or brittleness. Towards a systematic study of such issues, we introduce a synthetic framework for assessing the ability of VLMs to perform algorithmic visual reasoning, comprising three tasks: Table Readout, Grid Navigation, and Visual Analogy. Each has two levels of difficulty, SIMPLE and HARD, and even the SIMPLE versions are difficult for frontier VLMs. We propose strategies for training on the SIMPLE version of tasks that improve performance on the corresponding HARD task, i.e., simple-to-hard (S2H) generalization. This controlled setup, where each task also has an equivalent text-only version, allows a quantification of the modality imbalance and how it is impacted by training strategy. We show that 1) explicit image-to-text conversion is important in promoting S2H generalization on images, by transferring reasoning from text; 2) conversion can be internalized at test time. We also report results of mechanistic study of this phenomenon. We identify measures of gradient alignment that can identify training strategies that promote better S2H generalization. Ablations highlight the importance of chain-of-thought.