Scientific Research, vital for improving human life, is hindered by its inherent complexity, slow pace, and the need for specialized experts. To enhance its productivity, we propose a ResearchAgent, a large language model-powered research idea writing agent, which automatically generates problems, methods, and experiment designs while iteratively refining them based on scientific literature. Specifically, starting with a core paper as the primary focus to generate ideas, our ResearchAgent is augmented not only with relevant publications through connecting information over an academic graph but also entities retrieved from an entity-centric knowledge store based on their underlying concepts, mined and shared across numerous papers. In addition, mirroring the human approach to iteratively improving ideas with peer discussions, we leverage multiple ReviewingAgents that provide reviews and feedback iteratively. Further, they are instantiated with human preference-aligned large language models whose criteria for evaluation are derived from actual human judgments. We experimentally validate our ResearchAgent on scientific publications across multiple disciplines, showcasing its effectiveness in generating novel, clear, and valid research ideas based on human and model-based evaluation results.
We analyze how well pre-trained large language models (e.g., Llama2, GPT-4, Claude 3, etc) can do linear and non-linear regression when given in-context examples, without any additional training or gradient updates. Our findings reveal that several large language models (e.g., GPT-4, Claude 3) are able to perform regression tasks with a performance rivaling (or even outperforming) that of traditional supervised methods such as Random Forest, Bagging, or Gradient Boosting. For example, on the challenging Friedman #2 regression dataset, Claude 3 outperforms many supervised methods such as AdaBoost, SVM, Random Forest, KNN, or Gradient Boosting. We then investigate how well the performance of large language models scales with the number of in-context exemplars. We borrow from the notion of regret from online learning and empirically show that LLMs are capable of obtaining a sub-linear regret.
The integration of Artificial Intelligence (AI) techniques, particularly large language models (LLMs), in finance has garnered increasing academic attention. Despite progress, existing studies predominantly focus on tasks like financial text summarization, question-answering, and stock movement prediction (binary classification), the application of LLMs to financial risk prediction remains underexplored. Addressing this gap, in this paper, we introduce RiskLabs, a novel framework that leverages LLMs to analyze and predict financial risks. RiskLabs uniquely integrates multimodal financial data, including textual and vocal information from Earnings Conference Calls (ECCs), market-related time series data, and contextual news data to improve financial risk prediction. Empirical results demonstrate RiskLabs' effectiveness in forecasting both market volatility and variance. Through comparative experiments, we examine the contributions of different data sources to financial risk assessment and highlight the crucial role of LLMs in this process. We also discuss the challenges associated with using LLMs for financial risk prediction and explore the potential of combining them with multimodal data for this purpose.
Language models trained on internet-scale data sets have shown an impressive ability to solve problems in Natural Language Processing and Computer Vision. However, experience is showing that these models are frequently brittle in unexpected ways, and require significant scaffolding to ensure that they operate correctly in the larger systems that comprise "language-model agents." In this paper, we argue that behavior trees provide a unifying framework for combining language models with classical AI and traditional programming. We introduce Dendron, a Python library for programming language model agents using behavior trees. We demonstrate the approach embodied by Dendron in three case studies: building a chat agent, a camera-based infrastructure inspection agent for use on a mobile robot or vehicle, and an agent that has been built to satisfy safety constraints that it did not receive through instruction tuning or RLHF.
Portfolio management is one of the unresponded problems of the Motion Pictures Industry (MPI). To design an optimal portfolio for an MPI distributor, it is essential to predict the box office of each project. Moreover, for an accurate box office prediction, it is critical to consider the effect of the celebrities involved in each MPI project, which was impossible with any precedent expert-based method. Additionally, the asymmetric characteristic of MPI data decreases the performance of any predictive algorithm. In this paper, firstly, the fame score of the celebrities is determined using a large language model. Then, to tackle the asymmetric character of MPI's data, projects are classified. Furthermore, the box office prediction takes place for each class of projects. Finally, using a hybrid multi-attribute decision-making technique, the preferability of each project for the distributor is calculated, and benefiting from a bi-objective optimization model, the optimal portfolio is designed.
Recent advances in Automated Theorem Proving have shown the effectiveness of leveraging a (large) language model that generates tactics (i.e. proof steps) to search through proof states. The current model, while trained solely on successful proof paths, faces a discrepancy at the inference stage, as it must sample and try various tactics at each proof state until finding success, unlike its training which does not incorporate learning from failed attempts. Intuitively, a tactic that leads to a failed search path would indicate that similar tactics should receive less attention during the following trials. In this paper, we demonstrate the benefit of training models that additionally learn from failed search paths. Facing the lack of such trial-and-error data in existing open-source theorem-proving datasets, we curate a dataset on intuitionistic propositional logic theorems and formalize it in Lean, such that we can reliably check the correctness of proofs. We compare our model trained on relatively short trial-and-error information (TrialMaster) with models trained only on the correct paths and discover that the former solves more unseen theorems with lower trial searches.
As large language models (LLMs) have gained popularity for a variety of use cases, making them adaptable and controllable has become increasingly important, especially for user-facing applications. While the existing literature on LLM adaptation primarily focuses on finding a model (or models) that optimizes a single predefined objective, here we focus on the challenging case where the model must dynamically adapt to diverse -- and often changing -- user preferences. For this, we leverage adaptation methods based on linear weight interpolation, casting them as continuous multi-domain interpolators that produce models with specific prescribed generation characteristics on-the-fly. Specifically, we use low-rank updates to fine-tune a base model to various different domains, yielding a set of anchor models with distinct generation profiles. Then, we use the weight updates of these anchor models to parametrize the entire (infinite) class of models contained within their convex hull. We empirically show that varying the interpolation weights yields predictable and consistent change in the model outputs with respect to all of the controlled attributes. We find that there is little entanglement between most attributes and identify and discuss the pairs of attributes for which this is not the case. Our results suggest that linearly interpolating between the weights of fine-tuned models facilitates predictable, fine-grained control of model outputs with respect to multiple stylistic characteristics simultaneously.
The "AI for Science, Energy, and Security" report from DOE outlines a significant focus on developing and optimizing artificial intelligence workflows for a foundational impact on a broad range of DOE missions. With the pervasive usage of artificial intelligence (AI) and machine learning (ML) tools and techniques, their energy efficiency is likely to become the gating factor toward adoption. This is because generative AI (GenAI) models are massive energy hogs: for instance, training a 200-billion parameter large language model (LLM) at Amazon is estimated to have taken 11.9 GWh, which is enough to power more than a thousand average U.S. households for a year. Inference consumes even more energy, because a model trained once serve millions. Given this scale, high energy efficiency is key to addressing the power delivery problem of constructing and operating new supercomputers and datacenters specialized for AI workloads. In that regard, we outline software- and architecture-level research challenges and opportunities, setting the stage for creating cross-layer energy optimizations in AI systems.
Medical Report Grounding is pivotal in identifying the most relevant regions in medical images based on a given phrase query, a critical aspect in medical image analysis and radiological diagnosis. However, prevailing visual grounding approaches necessitate the manual extraction of key phrases from medical reports, imposing substantial burdens on both system efficiency and physicians. In this paper, we introduce a novel framework, Medical Report Grounding (MedRG), an end-to-end solution for utilizing a multi-modal Large Language Model to predict key phrase by incorporating a unique token, BOX, into the vocabulary to serve as an embedding for unlocking detection capabilities. Subsequently, the vision encoder-decoder jointly decodes the hidden embedding and the input medical image, generating the corresponding grounding box. The experimental results validate the effectiveness of MedRG, surpassing the performance of the existing state-of-the-art medical phrase grounding methods. This study represents a pioneering exploration of the medical report grounding task, marking the first-ever endeavor in this domain.
Some have criticised Generative AI Systems for replicating the familiar pathologies of already widely-deployed AI systems. Other critics highlight how they foreshadow vastly more powerful future systems, which might threaten humanity's survival. The first group says there is nothing new here; the other looks through the present to a perhaps distant horizon. In this paper, I instead pay attention to what makes these particular systems distinctive: both their remarkable scientific achievement, and the most likely and consequential ways in which they will change society over the next five to ten years. In particular, I explore the potential societal impacts and normative questions raised by the looming prospect of 'Language Model Agents', in which multimodal large language models (LLMs) form the executive centre of complex, tool-using AI systems that can take unsupervised sequences of actions towards some goal.
Objective. Vaccination has engendered a spectrum of public opinions, with social media acting as a crucial platform for health-related discussions. The emergence of artificial intelligence technologies, such as large language models (LLMs), offers a novel opportunity to efficiently investigate public discourses. This research assesses the accuracy of ChatGPT, a widely used and freely available service built upon an LLM, for sentiment analysis to discern different stances toward Human Papillomavirus (HPV) vaccination. Methods. Messages related to HPV vaccination were collected from social media supporting different message formats: Facebook (long format) and Twitter (short format). A selection of 1,000 human-evaluated messages was input into the LLM, which generated multiple response instances containing its classification results. Accuracy was measured for each message as the level of concurrence between human and machine decisions, ranging between 0 and 1. Results. Average accuracy was notably high when 20 response instances were used to determine the machine decision of each message: .882 (SE = .021) and .750 (SE = .029) for anti- and pro-vaccination long-form; .773 (SE = .027) and .723 (SE = .029) for anti- and pro-vaccination short-form, respectively. Using only three or even one instance did not lead to a severe decrease in accuracy. However, for long-form messages, the language model exhibited significantly lower accuracy in categorizing pro-vaccination messages than anti-vaccination ones. Conclusions. ChatGPT shows potential in analyzing public opinions on HPV vaccination using social media content. However, understanding the characteristics and limitations of a language model within specific public health contexts remains imperative.
Assessing the factual consistency of automatically generated texts in relation to source context is crucial for developing reliable natural language generation applications. Recent literature proposes AlignScore which uses a unified alignment model to evaluate factual consistency and substantially outperforms previous methods across many benchmark tasks. In this paper, we take a closer look of datasets used in AlignScore and uncover an unexpected finding: utilizing a smaller number of data points can actually improve performance. We process the original AlignScore training dataset to remove noise, augment with robustness-enhanced samples, and utilize a subset comprising 10\% of the data to train an improved factual consistency evaluation model, we call LIM-RA (Less Is More for Robust AlignScore). LIM-RA demonstrates superior performance, consistently outperforming AlignScore and other strong baselines like ChatGPT across four benchmarks (two utilizing traditional natural language generation datasets and two focused on large language model outputs). Our experiments show that LIM-RA achieves the highest score on 24 of the 33 test datasets, while staying competitive on the rest, establishing the new state-of-the-art benchmarks.
We address prevailing challenges of the brain-powered research, departing from the observation that the literature hardly recover accurate spatial information and require subject-specific models. To address these challenges, we propose UMBRAE, a unified multimodal decoding of brain signals. First, to extract instance-level conceptual and spatial details from neural signals, we introduce an efficient universal brain encoder for multimodal-brain alignment and recover object descriptions at multiple levels of granularity from subsequent multimodal large language model (MLLM). Second, we introduce a cross-subject training strategy mapping subject-specific features to a common feature space. This allows a model to be trained on multiple subjects without extra resources, even yielding superior results compared to subject-specific models. Further, we demonstrate this supports weakly-supervised adaptation to new subjects, with only a fraction of the total training data. Experiments demonstrate that UMBRAE not only achieves superior results in the newly introduced tasks but also outperforms methods in well established tasks. To assess our method, we construct and share with the community a comprehensive brain understanding benchmark BrainHub. Our code and benchmark are available at https://weihaox.github.io/UMBRAE.
This work introduces an efficient method to scale Transformer-based Large Language Models (LLMs) to infinitely long inputs with bounded memory and computation. A key component in our proposed approach is a new attention technique dubbed Infini-attention. The Infini-attention incorporates a compressive memory into the vanilla attention mechanism and builds in both masked local attention and long-term linear attention mechanisms in a single Transformer block. We demonstrate the effectiveness of our approach on long-context language modeling benchmarks, 1M sequence length passkey context block retrieval and 500K length book summarization tasks with 1B and 8B LLMs. Our approach introduces minimal bounded memory parameters and enables fast streaming inference for LLMs.
Large-scale high-quality training data is important for improving the performance of models. After trained with data that has rationales (reasoning steps), models gain reasoning capability. However, the dataset with high-quality rationales is relatively scarce due to the high annotation cost. To address this issue, we propose \textit{Self-motivated Learning} framework. The framework motivates the model itself to automatically generate rationales on existing datasets. Based on the inherent rank from correctness across multiple rationales, the model learns to generate better rationales, leading to higher reasoning capability. Specifically, we train a reward model with the rank to evaluate the quality of rationales, and improve the performance of reasoning through reinforcement learning. Experiment results of Llama2 7B on multiple reasoning datasets show that our method significantly improves the reasoning ability of models, even outperforming text-davinci-002 in some datasets.
This paper studies retrieval-augmented approaches for personalizing large language models (LLMs), which potentially have a substantial impact on various applications and domains. We propose the first attempt to optimize the retrieval models that deliver a limited number of personal documents to large language models for the purpose of personalized generation. We develop two optimization algorithms that solicit feedback from the downstream personalized generation tasks for retrieval optimization--one based on reinforcement learning whose reward function is defined using any arbitrary metric for personalized generation and another based on knowledge distillation from the downstream LLM to the retrieval model. This paper also introduces a pre- and post-generation retriever selection model that decides what retriever to choose for each LLM input. Extensive experiments on diverse tasks from the language model personalization (LaMP) benchmark reveal statistically significant improvements in six out of seven datasets.
The 2023 AI4EIC hackathon was the culmination of the third annual AI4EIC workshop at The Catholic University of America. This workshop brought together researchers from physics, data science and computer science to discuss the latest developments in Artificial Intelligence (AI) and Machine Learning (ML) for the Electron Ion Collider (EIC), including applications for detectors, accelerators, and experimental control. The hackathon, held on the final day of the workshop, involved using a chatbot powered by a Large Language Model, ChatGPT-3.5, to train a binary classifier neutrons and photons in simulated data from the \textsc{GlueX} Barrel Calorimeter. In total, six teams of up to four participants from all over the world took part in this intense educational and research event. This article highlights the hackathon challenge, the resources and methodology used, and the results and insights gained from analyzing physics data using the most cutting-edge tools in AI/ML.
The rapid advancement of artificial intelligence has led to significant improvements in automated decision-making. However, the increased performance of models often comes at the cost of explainability and transparency of their decision-making processes. In this paper, we investigate the capabilities of large language models to explain decisions, using football refereeing as a testing ground, given its decision complexity and subjectivity. We introduce the Explainable Video Assistant Referee System, X-VARS, a multi-modal large language model designed for understanding football videos from the point of view of a referee. X-VARS can perform a multitude of tasks, including video description, question answering, action recognition, and conducting meaningful conversations based on video content and in accordance with the Laws of the Game for football referees. We validate X-VARS on our novel dataset, SoccerNet-XFoul, which consists of more than 22k video-question-answer triplets annotated by over 70 experienced football referees. Our experiments and human study illustrate the impressive capabilities of X-VARS in interpreting complex football clips. Furthermore, we highlight the potential of X-VARS to reach human performance and support football referees in the future.
Recent advances in recurrent neural network architectures, such as Mamba and RWKV, have enabled RNNs to match or exceed the performance of equal-size transformers in terms of language modeling perplexity and downstream evaluations, suggesting that future systems may be built on completely new architectures. In this paper, we examine if selected interpretability methods originally designed for transformer language models will transfer to these up-and-coming recurrent architectures. Specifically, we focus on steering model outputs via contrastive activation addition, on eliciting latent predictions via the tuned lens, and eliciting latent knowledge from models fine-tuned to produce false outputs under certain conditions. Our results show that most of these techniques are effective when applied to RNNs, and we show that it is possible to improve some of them by taking advantage of RNNs' compressed state.
Echocardiography segmentation for cardiac analysis is time-consuming and resource-intensive due to the variability in image quality and the necessity to process scans from various standard views. While current automated segmentation methods in echocardiography show promising performance, they are trained on specific scan views to analyze corresponding data. However, this solution has a limitation as the number of required models increases with the number of standard views. To address this, in this paper, we present a prompt-driven universal method for view-agnostic echocardiography analysis. Considering the domain shift between standard views, we first introduce a method called prompt matching, aimed at learning prompts specific to different views by matching prompts and querying input embeddings using a pre-trained vision model. Then, we utilized a pre-trained medical language model to align textual information with pixel data for accurate segmentation. Extensive experiments on three standard views showed that our approach significantly outperforms the state-of-the-art universal methods and achieves comparable or even better performances over the segmentation model trained and tested on same views.