As language models (LMs) scale, they develop many novel behaviors, good and bad, exacerbating the need to evaluate how they behave. Prior work creates evaluations with crowdwork (which is time-consuming and expensive) or existing data sources (which are not always available). Here, we automatically generate evaluations with LMs. We explore approaches with varying amounts of human effort, from instructing LMs to write yes/no questions to making complex Winogender schemas with multiple stages of LM-based generation and filtering. Crowdworkers rate the examples as highly relevant and agree with 90-100% of labels, sometimes more so than corresponding human-written datasets. We generate 154 datasets and discover new cases of inverse scaling where LMs get worse with size. Larger LMs repeat back a dialog user's preferred answer ("sycophancy") and express greater desire to pursue concerning goals like resource acquisition and goal preservation. We also find some of the first examples of inverse scaling in RL from Human Feedback (RLHF), where more RLHF makes LMs worse. For example, RLHF makes LMs express stronger political views (on gun rights and immigration) and a greater desire to avoid shut down. Overall, LM-written evaluations are high-quality and let us quickly discover many novel LM behaviors.

Developing safe and useful general-purpose AI systems will require us to make progress on scalable oversight: the problem of supervising systems that potentially outperform us on most skills relevant to the task at hand. Empirical work on this problem is not straightforward, since we do not yet have systems that broadly exceed our abilities. This paper discusses one of the major ways we think about this problem, with a focus on how to turn it into one that can be productively studied empirically. We first present an experimental design centered on choosing tasks for which human specialists succeed but unaided humans and current general AI systems fail. We then present a proof-of-concept experiment following meant to demonstrate a key feature of this experimental design and show its viability with two question-answering tasks: MMLU and time-limited QuALITY. On these tasks, we find that human participants who interact with an unreliable large-language-model dialog assistant through chat -- a trivial baseline strategy for scalable oversight -- substantially outperform both the model alone and their own unaided performance. These results are an encouraging sign that scalable oversight will be tractable to study with present models and bolster recent findings that large language models can productively assist humans with difficult tasks.

In robotics and computer vision communities, extensive studies have been widely conducted regarding surveillance tasks, including human detection, tracking, and motion recognition with a camera. Additionally, deep learning algorithms are widely utilized in the aforementioned tasks as in other computer vision tasks. Existing public datasets are insufficient to develop learning-based methods that handle various surveillance for outdoor and extreme situations such as harsh weather and low illuminance conditions. Therefore, we introduce a new large-scale outdoor surveillance dataset named eXtremely large-scale Multi-modAl Sensor dataset (X-MAS) containing more than 500,000 image pairs and the first-person view data annotated by well-trained annotators. Moreover, a single pair contains multi-modal data (e.g. an IR image, an RGB image, a thermal image, a depth image, and a LiDAR scan). This is the first large-scale first-person view outdoor multi-modal dataset focusing on surveillance tasks to the best of our knowledge. We present an overview of the proposed dataset with statistics and present methods of exploiting our dataset with deep learning-based algorithms. The latest information on the dataset and our study are available at https://github.com/lge-robot-navi, and the dataset will be available for download through a server.

We introduce an end-to-end computational framework that enables hyperparameter optimization with the DeepHyper library, accelerated training, and interpretable AI inference with a suite of state-of-the-art AI models, including CGCNN, PhysNet, SchNet, MPNN, MPNN-transformer, and TorchMD-Net. We use these AI models and the benchmark QM9, hMOF, and MD17 datasets to showcase the prediction of user-specified materials properties in modern computing environments, and to demonstrate translational applications for the modeling of small molecules, crystals and metal organic frameworks with a unified, stand-alone framework. We deployed and tested this framework in the ThetaGPU supercomputer at the Argonne Leadership Computing Facility, and the Delta supercomputer at the National Center for Supercomputing Applications to provide researchers with modern tools to conduct accelerated AI-driven discovery in leadership class computing environments.

Sentence summarization shortens given texts while maintaining core contents of the texts. Unsupervised approaches have been studied to summarize texts without human-written summaries. However, recent unsupervised models are extractive, which remove words from texts and thus they are less flexible than abstractive summarization. In this work, we devise an abstractive model based on reinforcement learning without ground-truth summaries. We formulate the unsupervised summarization based on the Markov decision process with rewards representing the summary quality. To further enhance the summary quality, we develop a multi-summary learning mechanism that generates multiple summaries with varying lengths for a given text, while making the summaries mutually enhance each other. Experimental results show that the proposed model substantially outperforms both abstractive and extractive models, yet frequently generating new words not contained in input texts.

Detection Transformer (DETR) directly transforms queries to unique objects by using one-to-one bipartite matching during training and enables end-to-end object detection. Recently, these models have surpassed traditional detectors on COCO with undeniable elegance. However, they differ from traditional detectors in multiple designs, including model architecture and training schedules, and thus the effectiveness of one-to-one matching is not fully understood. In this work, we conduct a strict comparison between the one-to-one Hungarian matching in DETRs and the one-to-many label assignments in traditional detectors with non-maximum supervision (NMS). Surprisingly, we observe one-to-many assignments with NMS consistently outperform standard one-to-one matching under the same setting, with a significant gain of up to 2.5 mAP. Our detector that trains Deformable-DETR with traditional IoU-based label assignment achieved 50.2 COCO mAP within 12 epochs (1x schedule) with ResNet50 backbone, outperforming all existing traditional or transformer-based detectors in this setting. On multiple datasets, schedules, and architectures, we consistently show bipartite matching is unnecessary for performant detection transformers. Furthermore, we attribute the success of detection transformers to their expressive transformer architecture. Code is available at https://github.com/jozhang97/DETA.

Diffusion-based generative models have achieved remarkable success in image generation. Their guidance formulation allows an external model to plug-and-play control the generation process for various tasks without fine-tuning the diffusion model. However, the direct use of publicly available off-the-shelf models for guidance fails due to their poor performance on noisy inputs. For that, the existing practice is to fine-tune the guidance models with labeled data corrupted with noises. In this paper, we argue that this practice has limitations in two aspects: (1) performing on inputs with extremely various noises is too hard for a single model; (2) collecting labeled datasets hinders scaling up for various tasks. To tackle the limitations, we propose a novel strategy that leverages multiple experts where each expert is specialized in a particular noise range and guides the reverse process at its corresponding timesteps. However, as it is infeasible to manage multiple networks and utilize labeled data, we present a practical guidance framework termed Practical Plug-And-Play (PPAP), which leverages parameter-efficient fine-tuning and data-free knowledge transfer. We exhaustively conduct ImageNet class conditional generation experiments to show that our method can successfully guide diffusion with small trainable parameters and no labeled data. Finally, we show that image classifiers, depth estimators, and semantic segmentation models can guide publicly available GLIDE through our framework in a plug-and-play manner.

Federated Learning has emerged to cope with raising concerns about privacy breaches in using Machine or Deep Learning models. This new paradigm allows the leverage of deep learning models in a distributed manner, enhancing privacy preservation. However, the server's blindness to local datasets introduces its vulnerability to model poisoning attacks and data heterogeneity, tampering with the global model performance. Numerous works have proposed robust aggregation algorithms and defensive mechanisms, but the approaches are orthogonal to individual attacks or issues. FedCC, the proposed method, provides robust aggregation by comparing the Centered Kernel Alignment of Penultimate Layers Representations. The experiment results on FedCC demonstrate that it mitigates untargeted and targeted model poisoning or backdoor attacks while also being effective in non-Independently and Identically Distributed data environments. By applying FedCC against untargeted attacks, global model accuracy is recovered the most. Against targeted backdoor attacks, FedCC nullified attack confidence while preserving the test accuracy. Most of the experiment results outstand the baseline methods.

Routine clinical visits of a patient produce not only image data, but also non-image data containing clinical information regarding the patient, i.e., medical data is multi-modal in nature. Such heterogeneous modalities offer different and complementary perspectives on the same patient, resulting in more accurate clinical decisions when they are properly combined. However, despite its significance, how to effectively fuse the multi-modal medical data into a unified framework has received relatively little attention. In this paper, we propose an effective graph-based framework called HetMed (Heterogeneous Graph Learning for Multi-modal Medical Data Analysis) for fusing the multi-modal medical data. Specifically, we construct a multiplex network that incorporates multiple types of non-image features of patients to capture the complex relationship between patients in a systematic way, which leads to more accurate clinical decisions. Extensive experiments on various real-world datasets demonstrate the superiority and practicality of HetMed. The source code for HetMed is available at https://github.com/Sein-Kim/Multimodal-Medical.

This work presents six structural quality metrics that can measure the quality of knowledge graphs and analyzes five cross-domain knowledge graphs on the web (Wikidata, DBpedia, YAGO, Google Knowledge Graph, Freebase) as well as 'Raftel', Naver's integrated knowledge graph. The 'Good Knowledge Graph' should define detailed classes and properties in its ontology so that knowledge in the real world can be expressed abundantly. Also, instances and RDF triples should use the classes and properties actively. Therefore, we tried to examine the internal quality of knowledge graphs numerically by focusing on the structure of the ontology, which is the schema of knowledge graphs, and the degree of use thereof. As a result of the analysis, it was possible to find the characteristics of a knowledge graph that could not be known only by scale-related indicators such as the number of classes and properties.