Visual language tasks such as Visual Question Answering (VQA) or Visual Entailment (VE) require AI models to comprehend and reason with both visual and textual content. Driven by the power of Large Language Models (LLMs), two prominent methods have emerged: (1) the hybrid integration between LLMs and Vision-Language Models (VLMs), where visual inputs are firstly converted into language descriptions by VLMs, serving as inputs for LLMs to generate final answer(s); (2) visual feature alignment in language space, where visual inputs are encoded as embeddings and projected to LLMs' language space via further supervised fine-tuning. The first approach provides light training costs and interpretability but is hard to be optimized in an end-to-end fashion. The second approach presents decent performance, but feature alignment usually requires large amounts of training data and lacks interpretability. To tackle this dilemma, we propose a novel approach, Inner Monologue Multi-Modal Optimization (IMMO), to solve complex vision language problems by simulating Inner Monologue, a cognitive process in which an individual engages in silent verbal communication with themselves. More specifically, we enable LLMs and VLMs to interact through natural language conversation (i.e., Inner Monologue) and propose to use a two-stage training process to learn how to do Inner Monologue (self-asking questions and answering questions). IMMO is evaluated on two popular tasks and achieves competitive performance with less training data when compared with state-of-the-art models while concurrently keeping the interpretability. The results suggest that by emulating the cognitive phenomenon of internal dialogue, our approach can enhance reasoning and explanation abilities, contributing to the more effective fusion of vision and language models. More importantly, instead of using predefined human-crafted monologues, IMMO learns this process within the deep learning models, broadening its potential applications across various AI challenges beyond vision and language tasks.

Large language models are increasingly becoming a cornerstone technology in artificial intelligence, the sciences, and society as a whole, yet the optimal strategies for dataset composition and filtering remain largely elusive. Many of the top-performing models lack transparency in their dataset curation and model development processes, posing an obstacle to the development of fully open language models. In this paper, we identify three core data-related challenges that must be addressed to advance open-source language models. These include (1) transparency in model development, including the data curation process, (2) access to large quantities of high-quality data, and (3) availability of artifacts and metadata for dataset curation and analysis. To address these challenges, we release RedPajama-V1, an open reproduction of the LLaMA training dataset. In addition, we release RedPajama-V2, a massive web-only dataset consisting of raw, unfiltered text data together with quality signals and metadata.Together, the RedPajama datasets comprise over 100 trillion tokens spanning multiple domains and with their quality signals facilitate the filtering of data, aiming to inspire the development of numerous new datasets. To date, these datasets have already been used in the training of strong language models used in production, such as Snowflake Arctic, Salesforce's XGen and AI2's OLMo. To provide insight into the quality of RedPajama, we present a series of analyses and ablation studies with decoder-only language models with up to 1.6B parameters. Our findings demonstrate how quality signals for web data can be effectively leveraged to curate high-quality subsets of the dataset, underscoring the potential of RedPajama to advance the development of transparent and high-performing language models at scale.

Although the Retrieval-Augmented Generation (RAG) paradigms can use external knowledge to enhance and ground the outputs of Large Language Models (LLMs) to mitigate generative hallucinations and static knowledge base problems, they still suffer from limited flexibility in adopting Information Retrieval (IR) systems with varying capabilities, constrained interpretability during the multi-round retrieval process, and a lack of end-to-end optimization. To address these challenges, we propose a novel LLM-centric approach, IM-RAG, that integrates IR systems with LLMs to support multi-round RAG through learning Inner Monologues (IM, i.e., the human inner voice that narrates one's thoughts). During the IM process, the LLM serves as the core reasoning model (i.e., Reasoner ) to either propose queries to collect more information via the Retriever or to provide a final answer based on the conversational context. We also introduce a Refiner that improves the outputs from the Retriever, effectively bridging the gap between the Reasoner and IR modules with varying capabilities and fostering multi-round communications. The entire IM process is optimized via Reinforcement Learning (RL) where a Progress Tracker is incorporated to provide mid-step rewards, and the answer prediction is further separately optimized via Supervised Fine-Tuning (SFT). We conduct extensive experiments with the HotPotQA dataset, a popular benchmark for retrieval-based, multi-step question-answering. The results show that our approach achieves state-of-the-art (SOTA) performance while providing high flexibility in integrating IR modules as well as strong interpretability exhibited in the learned inner monologue.

Sketch recognition has been studied for decades, but it is far from solved. Drawing styles are highly variable across people and adapting to idiosyncratic visual expressions requires data-efficient learning. Explainability also matters, so that users can see why a system got confused about something. This paper introduces a novel part-based approach for sketch recognition, based on hierarchical analogical learning, a new method to apply analogical learning to qualitative representations. Given a sketched object, our system automatically segments it into parts and constructs multi-level qualitative representations of them. Our approach performs analogical generalization at multiple levels of part descriptions and uses coarse-grained results to guide interpretation at finer levels. Experiments on the Berlin TU dataset and the Coloring Book Objects dataset show that the system can learn explainable models in a data-efficient manner.

Visual Relation Detection is currently one of the most popular problems for visual understanding. Many deep-learning models are designed for relation detection on images and have achieved impressive results. However, deep-learning models have several serious problems, including poor training-efficiency and lack of understandability. Psychologists have ample evidence that analogy is central in human learning and reasoning, including visual reasoning. This paper introduces a new hybrid system for visual relation detection combining deep-learning models and analogical generalization. Object bounding boxes and masks are detected using deep-learning models and analogical generalization over qualitative representations is used for visual relation detection between object pairs. Experiments on the Visual Relation Detection dataset indicates that our hybrid system gets comparable results on the task and is more training-efficient and explainable than pure deep-learning models.

Generating formal-language programs represented by relational tuples, such as Lisp programs or mathematical operations, to solve problems stated in natural language is a challenging task because it requires explicitly capturing discrete symbolic structural information implicit in the input. However, most general neural sequence models do not explicitly capture such structural information, limiting their performance on these tasks. In this paper, we propose a new encoder-decoder model based on a structured neural representation, Tensor Product Representations (TPRs), for mapping Natural-language problems to Formal-language solutions, called TP-N2F. The encoder of TP-N2F employs TPR ‘binding’ to encode natural-language symbolic structure in vector space and the decoder uses TPR ‘unbinding’ to generate, in symbolic space, a sequential program represented by relational tuples, each consisting of a relation (or operation) and a number of arguments. TP-N2F considerably outperforms LSTM-based seq2seq models on two benchmarks and creates new state-of-the-art results. Ablation studies show that improvements can be attributed to the use of structured TPRs explicitly in both the encoder and decoder. Analysis of the learned structures shows how TPRs enhance the interpretability of TP-N2F.

Deep learning systems can perform well on some image recognition tasks. However, they have serious limitations, including requiring far more training data than humans do and being fooled by adversarial examples. By contrast, analogical learning over relational representations tends to be far more data-efficient, requiring only human-like amounts of training data. This paper introduces an approach that combines automatically constructed qualitative visual representations with analogical learning to tackle a hard computer vision problem, object recognition from sketches. Results from the MNIST dataset and a novel dataset, the Coloring Book Objects dataset, are provided. Comparison to existing approaches indicates that analogical generalization can be used to identify sketched objects from these datasets with several orders of magnitude fewer examples than deep learning systems require.

Human action recognition remains a difficult problem for AI. Traditional machine learning techniques can have high recognition accuracy, but they are typically black boxes whose internal models are not inspectable and whose results are not explainable. This paper describes a new pipeline for recognizing human actions from skeleton data via analogical generalization. Specifically, starting with Kinect data, we segment each human action by temporal regions where the motion is qualitatively uniform, creating a sketch graph that provides a form of qualitative representation of the behavior that is easy to visualize. Models are learned from sketch graphs via analogical generalization, which are then used for classification via analogical retrieval. The retrieval process also produces links between the new example and components of the model that provide explanations. To improve recognition accuracy, we implement dynamic feature selection to pick reasonable relational features. We show the explanation advantage of our approach by example, and results on three public datasets illustrate its utility.

Many individuals exhibit unconscious body movements called mannerisms while speaking. These repeated changes often distract the audience when not relevant to the verbal context. We present an intelligent interface that can automatically extract human gestures using Microsoft Kinect to make speakers aware of their mannerisms. We use a sparsity-based algorithm, Shift Invariant Sparse Coding, to automatically extract the patterns of body movements. These patterns are displayed in an interface with subtle question and answer-based feedback scheme that draws attention to the speaker's body language. Our formal evaluation with 27 participants shows that the users became aware of their body language after using the system. In addition, when independent observers annotated the accuracy of the algorithm for every extracted pattern, we find that the patterns extracted by our algorithm is significantly (p<0.001) more accurate than just random selection. This represents a strong evidence that the algorithm is able to extract human-interpretable body movement patterns. An interactive demo of AutoManner is available at http://tinyurl.com/AutoManner.

Fashion is a perpetual topic in human social life, and the mass has the penchant to emulate what large city residents and celebrities wear. Undeniably, New York City is such a bellwether large city with all kinds of fashion leadership. Consequently, to study what the fashion trends are during this year, it is very helpful to learn the fashion trends of New York City. Discovering fashion trends in New York City could boost many applications such as clothing recommendation and advertising. Does the fashion trend in the New York Fashion Show actually influence the clothing styles on the public? To answer this question, we design a novel system that consists of three major components: (1) constructing a large dataset from the New York Fashion Shows and New York street chic in order to understand the likely clothing fashion trends in New York, (2) utilizing a learning-based approach to discover fashion attributes as the representative characteristics of fashion trends, and (3) comparing the analysis results from the New York Fashion Shows and street-chic images to verify whether the fashion shows have actual influence on the people in New York City. Through the preliminary experiments over a large clothing dataset, we demonstrate the effectiveness of our proposed system, and obtain useful insights on fashion trends and fashion influence.