We propose a method to fuse frozen text-only large language models (LLMs) with pre-trained image encoder and decoder models, by mapping between their embedding spaces. Our model demonstrates a wide suite of multimodal capabilities: image retrieval, novel image generation, and multimodal dialogue. Ours is the first approach capable of conditioning on arbitrarily interleaved image and text inputs to generate coherent image (and text) outputs. To achieve strong performance on image generation, we propose an efficient mapping network to ground the LLM to an off-the-shelf text-to-image generation model. This mapping network translates hidden representations of text into the embedding space of the visual models, enabling us to leverage the strong text representations of the LLM for visual outputs.

Modern large language models become multimodal, analyzing various data formats like text and images. While fine-tuning is effective for adapting these multimodal language models (MLMs) to downstream tasks, full fine-tuning is computationally expensive. Parameter-Efficient Fine-Tuning (PEFT) methods address this by training only a small portion of model weights. However, MLMs are difficult to interpret, making it challenging to identify which components are most effective for training to balance efficiency and performance. We propose an attention-based interpretability method for MLMs by analyzing attention scores relative to image tokens. The core idea is to identify attention heads that focus on image key objects. We utilize this information to select optimal model components for PEFT in multimodal models. Our contributions include a method for identifying attention heads associated with image key objects, its application to PEFT for image captioning, and the creation of a new dataset containing images, key object masks, and their textual descriptions. We conducted experiments on MLMs with 2-3 billion parameters to validate the method's effectiveness. By calculating Head Impact (HI) scores we quantify an attention head's focus on key objects, indicating its significance in image understanding. Our fine-tuning experiments demonstrate that adapting layers with the highest HI scores leads to the most significant shifts in metrics compared to pre-trained, randomly selected, or lowest-HI-score layers. This indicates that fine-tuning a small percentage (around 0.01%) of parameters in these crucial layers can substantially influence image understanding capabilities.

Multimodal language models (MLMs) show promise for clinical decision support and diagnostic reasoning, raising the prospect of end-to-end automated medical image interpretation. However, clinicians are highly selective in adopting AI tools; a model that makes errors on seemingly simple perception tasks such as determining image orientation or identifying whether a CT scan is contrast-enhance are unlikely to be adopted for clinical tasks. We introduce Medblink, a benchmark designed to probe these models for such perceptual abilities. Medblink spans eight clinically meaningful tasks across multiple imaging modalities and anatomical regions, totaling 1,429 multiple-choice questions over 1,605 images. We evaluate 19 state-of-the-art MLMs, including general purpose (GPT4o, Claude 3.5 Sonnet) and domain specific (Med Flamingo, LLaVA Med, RadFM) models. While human annotators achieve 96.4% accuracy, the best-performing model reaches only 65%. These results show that current MLMs frequently fail at routine perceptual checks, suggesting the need to strengthen their visual grounding to support clinical adoption. Data is available on our project page.

Personal service robots are deployed to support daily living in domestic environments, particularly for elderly and individuals requiring assistance. These robots must perceive complex and dynamic surroundings, understand tasks, and execute context-appropriate actions. However, current systems rely on proprietary, hard-coded solutions tied to specific hardware and software, resulting in siloed implementations that are difficult to adapt and scale across platforms. Ontologies and Knowledge Graphs (KGs) offer a solution to enable interoperability across systems, through structured and standardized representations of knowledge and reasoning. However, symbolic systems such as KGs and ontologies struggle with raw and noisy sensory input. In contrast, multimodal language models are well suited for interpreting input such as images and natural language, but often lack transparency, consistency, and knowledge grounding. In this work, we propose a neurosymbolic framework that combines the perceptual strengths of multimodal language models with the structured representations provided by KGs and ontologies, with the aim of supporting interoperability in robotic applications. Our approach generates ontology-compliant KGs that can inform robot behavior in a platform-independent manner. We evaluated this framework by integrating robot perception data, ontologies, and five multimodal models (three LLaMA and two GPT models), using different modes of neural-symbolic interaction. We assess the consistency and effectiveness of the generated KGs across multiple runs and configurations, and perform statistical analyzes to evaluate performance. Results show that GPT-o1 and LLaMA 4 Maverick consistently outperform other models. However, our findings also indicate that newer models do not guarantee better results, highlighting the critical role of the integration strategy in generating ontology-compliant KGs.

Humans draw to facilitate reasoning: we draw auxiliary lines when solving geometry problems; we mark and circle when reasoning on maps; we use sketches to amplify our ideas and relieve our limited-capacity working memory. However, such actions are missing in current multimodal language models (LMs). Current chain-of-thought and tool-use paradigms only use text as intermediate reasoning steps. In this work, we introduce Sketchpad, a framework that gives multimodal LMs a visual sketchpad and tools to draw on the sketchpad. The LM conducts planning and reasoning according to the visual artifacts it has drawn.

We propose a method to fuse frozen text-only large language models (LLMs) with pre-trained image encoder and decoder models, by mapping between their embedding spaces. Our model demonstrates a wide suite of multimodal capabilities: image retrieval, novel image generation, and multimodal dialogue. Ours is the first approach capable of conditioning on arbitrarily interleaved image and text inputs to generate coherent image (and text) outputs. To achieve strong performance on image generation, we propose an efficient mapping network to ground the LLM to an off-the-shelf text-to-image generation model. This mapping network translates hidden representations of text into the embedding space of the visual models, enabling us to leverage the strong text representations of the LLM for visual outputs.