Training vision language models (VLMs) aims to align visual representations from a vision encoder with the textual representations of a pretrained large language model (LLM). However, many VLMs exhibit reduced factual recall performance compared to their LLM backbones, raising the question of how effective multimodal fine-tuning is at extending existing mechanisms within the LLM to visual inputs. We argue that factual recall based on visual inputs requires VLMs to solve a two-hop problem: (1) forming entity representations from visual inputs, and (2) recalling associated factual knowledge based on these entity representations. By benchmarking 14 VLMs with various architectures (LLaVA, Native, Cross-Attention), sizes (7B-124B parameters), and training setups on factual recall tasks against their original LLM backbone models, we find that 11 of 14 models exhibit factual recall degradation. We select three models exhibiting high-and two models with low performance degradation, and use attribution patching, activation patching, and probing to show that degraded VLMs struggle to use the existing factual recall circuit of their LLM backbone, because they resolve the first hop too late in the computation. In contrast, high-performing VLMs resolve entity representations early enough to reuse the existing factual recall mechanism. Finally, we demonstrate two methods to recover performance: patching entity representations from the LLM backbone into the VLM, and prompting with chain-of-thought reasoning. Our results highlight that the speed of early entity resolution critically determines how effective VLMs are in using preexisting LLM mechanisms. More broadly, our work illustrates how mechanistic analysis can explain and unveil systematic failures in multimodal alignment.

Large Vision-Language Models (LVLMs) have made substantial progress by integrating pre-trained large language models (LLMs) and vision models through instruction tuning. Despite these advancements, LVLMs often exhibit the hallucination phenomenon, where generated text responses appear linguistically plausible but contradict the input image, indicating a misalignment between image and text pairs. This misalignment arises because the model tends to prioritize textual information over visual input, even when both the language model and visual representations are of high quality. Existing methods leverage additional models or human annotations to curate preference data and enhance modality alignment through preference optimization. These approaches are resource-intensive and may not effectively reflect the target LVLM's preferences, making the curated preferences easily distinguishable.

Co-first author order is arbitrary and may be swapped when citing this work.

We propose a method for knowledge transfer based on a regularization term in our loss function, supervising the sequence of required reasoning operations.

Human visual reasoning is characterized by an ability to identify abstract patterns from only a small number of examples, and to systematically generalize those patterns to novel inputs. This capacity depends in large part on our ability to represent complex visual inputs in terms of both objects and relations. Recent work in computer vision has introduced models with the capacity to extract object-centric representations, leading to the ability to process multi-object visual inputs, but falling short of the systematic generalization displayed by human reasoning. Other recent models have employed inductive biases for relational abstraction to achieve systematic generalization of learned abstract rules, but have generally assumed the presence of object-focused inputs. Here, we combine these two approaches, introducing Object-Centric Relational Abstraction (OCRA), a model that extracts explicit representations of both objects and abstract relations, and achieves strong systematic generalization in tasks (including a novel dataset, CLEVR-ART, with greater visual complexity) involving complex visual displays.

Infants possess a remarkable ability to rapidly learn and process visual inputs. As an infant's mobility increases, so does the variety and dynamics of their visual inputs. Is this change in the properties of the visual inputs beneficial or even critical for the proper development of the visual system? To address this question, we used video recordings from infants wearing head-mounted cameras to train a variety of self-supervised learning models. Critically, we separated the infant data by age group and evaluated the importance of training with a curriculum aligned with developmental order. We found that initiating learning with the data from the youngest age group provided the strongest learning signal and led to the best learning outcomes in terms of downstream task performance. We then showed that the benefits of the data from the youngest age group are due to the slowness and simplicity of the visual experience. The results provide strong empirical evidence for the importance of the properties of the early infant experience and developmental progression in training. More broadly, our approach and findings take a noteworthy step towards reverse engineering the learning mechanisms in newborn brains using image-computable models from artificial intelligence.

Humans learn from visual inputs at multiple timescales, both rapidly and flexibly acquiring visual knowledge over short periods, and robustly accumulating online learning progress over longer periods. Modeling these powerful learning capabilities is an important problem for computational visual cognitive science, and models that could replicate them would be of substantial utility in real-world computer vision settings. In this work, we establish benchmarks for both real-time and life-long continual visual learning. Our real-time learning benchmark measures a model's ability to match the rapid visual behavior changes of real humans over the course of minutes and hours, given a stream of visual inputs. Our life-long learning benchmark evaluates the performance of models in a purely online learning curriculum obtained directly from child visual experience over the course of years of development.