Training Small Multimodal Models to Bridge Biomedical Competency Gap: A Case Study in Radiology Imaging

The scaling laws and extraordinary performance of large foundation models motivate the development and utilization of such large models in biomedicine. However, despite early promising results on some biomedical benchmarks, there are still major challenges that need to be addressed before these models can be used in real-world applications. Frontier models such as GPT-4V still have major competency gaps in multimodal capabilities for biomedical applications. Moreover, pragmatic issues such as access, cost, latency, and compliance make it hard for clinicians to use privately-hosted state-of-the-art large models directly on private patient data. In this paper, we explore training open-source small multimodal models (SMMs) to bridge biomedical competency gaps for unmet clinical needs. To maximize data efficiency, we adopt a modular approach by incorporating state-of-the-art pre-trained models for image and text modalities, and focusing on training a lightweight adapter to ground each modality to the text embedding space. We conduct a comprehensive study of this approach on radiology imaging. For training, we assemble a large dataset with over 1 million image-text pairs. For evaluation, we propose a clinically driven novel approach using GPT-4 and demonstrate its parity with expert evaluation. We also study grounding qualitatively using attention. For best practice, we conduct a systematic ablation study on various choices in data engineering and multimodal training. The resulting LLaVA-Rad (7B) model attains state-of-the-art results on radiology tasks such as report generation and cross-modal retrieval, even outperforming much larger models such as GPT-4V and Med-PaLM M (84B). LLaVA-Rad is fast and can be run on a single V100 GPU in private settings, offering a promising state-of-the-art tool for real-world clinical applications.

CoDi-2: In-Context, Interleaved, and Interactive Any-to-Any Generation

We present CoDi-2, a versatile and interactive Multimodal Large Language Model (MLLM) that can follow complex multimodal interleaved instructions, conduct in-context learning (ICL), reason, chat, edit, etc., in an any-to-any input-output modality paradigm. By aligning modalities with language for both encoding and generation, CoDi-2 empowers Large Language Models (LLMs) to not only understand complex modality-interleaved instructions and in-context examples, but also autoregressively generate grounded and coherent multimodal outputs in the continuous feature space. To train CoDi-2, we build a large-scale generation dataset encompassing in-context multimodal instructions across text, vision, and audio. CoDi-2 demonstrates a wide range of zero-shot capabilities for multimodal generation, such as in-context learning, reasoning, and compositionality of any-to-any modality generation through multi-round interactive conversation. CoDi-2 surpasses previous domain-specific models on tasks such as subject-driven image generation, vision transformation, and audio editing. CoDi-2 signifies a substantial breakthrough in developing a comprehensive multimodal foundation model adept at interpreting in-context language-vision-audio interleaved instructions and producing multimodal outputs.

i-Code Studio: A Configurable and Composable Framework for Integrative AI

Artificial General Intelligence (AGI) requires comprehensive understanding and generation capabilities for a variety of tasks spanning different modalities and functionalities. Integrative AI is one important direction to approach AGI, through combining multiple models to tackle complex multimodal tasks. However, there is a lack of a flexible and composable platform to facilitate efficient and effective model composition and coordination. In this paper, we propose the i-Code Studio, a configurable and composable framework for Integrative AI. The i-Code Studio orchestrates multiple pre-trained models in a finetuning-free fashion to conduct complex multimodal tasks. Instead of simple model composition, the i-Code Studio provides an integrative, flexible, and composable setting for developers to quickly and easily compose cutting-edge services and technologies tailored to their specific requirements. The i-Code Studio achieves impressive results on a variety of zero-shot multimodal tasks, such as video-to-text retrieval, speech-to-speech translation, and visual question answering. We also demonstrate how to quickly build a multimodal agent based on the i-Code Studio that can communicate and personalize for users.

i-Code V2: An Autoregressive Generation Framework over Vision, Language, and Speech Data

The convergence of text, visual, and audio data is a key step towards human-like artificial intelligence, however the current Vision-Language-Speech landscape is dominated by encoder-only models which lack generative abilities. We propose closing this gap with i-Code V2, the first model capable of generating natural language from any combination of Vision, Language, and Speech data. i-Code V2 is an integrative system that leverages state-of-the-art single-modality encoders, combining their outputs with a new modality-fusing encoder in order to flexibly project combinations of modalities into a shared representational space. Next, language tokens are generated from these representations via an autoregressive decoder. The whole framework is pretrained end-to-end on a large collection of dual- and single-modality datasets using a novel text completion objective that can be generalized across arbitrary combinations of modalities. i-Code V2 matches or outperforms state-of-the-art single- and dual-modality baselines on 7 multimodal tasks, demonstrating the power of generative multimodal pretraining across a diversity of tasks and signals.

Learning to Represent Programs with Graphs

Learning tasks on source code (i.e., formal languages) have been considered recently, but most work has tried to transfer natural language methods and does not capitalize on the unique opportunities offered by code's known syntax. For example, long-range dependencies induced by using the same variable or function in distant locations are often not considered. We propose to use graphs to represent both the syntactic and semantic structure of code and use graph-based deep learning methods to learn to reason over program structures. In this work, we present how to construct graphs from source code and how to scale Gated Graph Neural Networks training to such large graphs. We evaluate our method on two tasks: VarNaming, in which a network attempts to predict the name of a variable given its usage, and VarMisuse, in which the network learns to reason about selecting the correct variable that should be used at a given program location. Our comparison to methods that use less structured program representations shows the advantages of modeling known structure, and suggests that our models learn to infer meaningful names and to solve the VarMisuse task in many cases. Additionally, our testing showed that VarMisuse identifies a number of bugs in mature open-source projects.

Relative Facial Action Unit Detection

This paper presents a subject-independent facial action unit (AU) detection method by introducing the concept of relative AU detection, for scenarios where the neutral face is not provided. We propose a new classification objective function which analyzes the temporal neighborhood of the current frame to decide if the expression recently increased, decreased or showed no change. This approach is a significant change from the conventional absolute method which decides about AU classification using the current frame, without an explicit comparison with its neighboring frames. Our proposed method improves robustness to individual differences such as face scale and shape, age-related wrinkles, and transitions among expressions (e.g., lower intensity of expressions). Our experiments on three publicly available datasets (Extended Cohn-Kanade (CK+), Bosphorus, and DISFA databases) show significant improvement of our approach over conventional absolute techniques. Keywords: facial action coding system (FACS); relative facial action unit detection; temporal information;

Facial Expression Representation and Recognition Using 2DHLDA, Gabor Wavelets, and Ensemble Learning

In this paper, a novel method for representation and recognition of the facial expressions in two-dimensional image sequences is presented. We apply a variation of two-dimensional heteroscedastic linear discriminant analysis (2DHLDA) algorithm, as an efficient dimensionality reduction technique, to Gabor representation of the input sequence. 2DHLDA is an extension of the two-dimensional linear discriminant analysis (2DLDA) approach and it removes the equal within-class covariance. By applying 2DHLDA in two directions, we eliminate the correlations between both image columns and image rows. Then, we perform a one-dimensional LDA on the new features. This combined method can alleviate the small sample size problem and instability encountered by HLDA. Also, employing both geometric and appearance features and using an ensemble learning scheme based on data fusion, we create a classifier which can efficiently classify the facial expressions. The proposed method is robust to illumination changes and it can properly represent temporal information as well as subtle changes in facial muscles. We provide experiments on Cohn-Kanade database that show the superiority of the proposed method. KEYWORDS: two-dimensional heteroscedastic linear discriminant analysis (2DHLDA), subspace learning, facial expression analysis, Gabor wavelets, ensemble learning.

Local Component Analysis for Nonparametric Bayes Classifier

The decision boundaries of Bayes classifier are optimal because they lead to maximum probability of correct decision. It means if we knew the prior probabilities and the class-conditional densities, we could design a classifier which gives the lowest probability of error. However, in classification based on nonparametric density estimation methods such as Parzen windows, the decision regions depend on the choice of parameters such as window width. Moreover, these methods suffer from curse of dimensionality of the feature space and small sample size problem which severely restricts their practical applications. In this paper, we address these problems by introducing a novel dimension reduction and classification method based on local component analysis. In this method, by adopting an iterative cross-validation algorithm, we simultaneously estimate the optimal transformation matrices (for dimension reduction) and classifier parameters based on local information. The proposed method can classify the data with complicated boundary and also alleviate the course of dimensionality dilemma. Experiments on real data show the superiority of the proposed algorithm in term of classification accuracies for pattern classification applications like age, facial expression and character recognition. Keywords: Bayes classifier, curse of dimensionality dilemma, Parzen window, pattern classification, subspace learning.

New S-norm and T-norm Operators for Active Learning Method

Active Learning Method (ALM) is a soft computing method used for modeling and control based on fuzzy logic. All operators defined for fuzzy sets must serve as either fuzzy S-norm or fuzzy T-norm. Despite being a powerful modeling method, ALM does not possess operators which serve as S-norms and T-norms which deprive it of a profound analytical expression/form. This paper introduces two new operators based on morphology which satisfy the following conditions: First, they serve as fuzzy S-norm and T-norm. Second, they satisfy Demorgans law, so they complement each other perfectly. These operators are investigated via three viewpoints: Mathematics, Geometry and fuzzy logic.

Extended Active Learning Method

Active Learning Method (ALM) is a soft computing method which is used for modeling and control, based on fuzzy logic. Although ALM has shown that it acts well in dynamic environments, its operators cannot support it very well in complex situations due to losing data. Thus ALM can find better membership functions if more appropriate operators be chosen for it. This paper substituted two new operators instead of ALM original ones; which consequently renewed finding membership functions in a way superior to conventional ALM. This new method is called Extended Active Learning Method (EALM).