Minsuk Chang

Sanghyuk Chun, Wonjae Kim, Song Park et al.

Image-Text matching (ITM) is a common task for evaluating the quality of Vision and Language (VL) models. However, existing ITM benchmarks have a significant limitation. They have many missing correspondences, originating from the data construction process itself. For example, a caption is only matched with one image although the caption can be matched with other similar images and vice versa. To correct the massive false negatives, we construct the Extended COCO Validation (ECCV) Caption dataset by supplying the missing associations with machine and human annotators. We employ five state-of-the-art ITM models with diverse properties for our annotation process. Our dataset provides x3.6 positive image-to-caption associations and x8.5 caption-to-image associations compared to the original MS-COCO. We also propose to use an informative ranking-based metric mAP@R, rather than the popular Recall@K (R@K). We re-evaluate the existing 25 VL models on existing and proposed benchmarks. Our findings are that the existing benchmarks, such as COCO 1K R@K, COCO 5K R@K, CxC R@1 are highly correlated with each other, while the rankings change when we shift to the ECCV mAP@R. Lastly, we delve into the effect of the bias introduced by the choice of machine annotator. Source code and dataset are available at https://github.com/naver-ai/eccv-caption

Dongyoon Han, Junsuk Choe, Seonghyeok Chun et al.

Supervised learning of image classifiers distills human knowledge into a parametric model through pairs of images and corresponding labels (X,Y). We argue that this simple and widely used representation of human knowledge neglects rich auxiliary information from the annotation procedure, such as the time-series of mouse traces and clicks left after image selection. Our insight is that such annotation byproducts Z provide approximate human attention that weakly guides the model to focus on the foreground cues, reducing spurious correlations and discouraging shortcut learning. To verify this, we create ImageNet-AB and COCO-AB. They are ImageNet and COCO training sets enriched with sample-wise annotation byproducts, collected by replicating the respective original annotation tasks. We refer to the new paradigm of training models with annotation byproducts as learning using annotation byproducts (LUAB). We show that a simple multitask loss for regressing Z together with Y already improves the generalisability and robustness of the learned models. Compared to the original supervised learning, LUAB does not require extra annotation costs. ImageNet-AB and COCO-AB are at https://github.com/naver-ai/NeglectedFreeLunch.

Jesse Zhang, Jiahui Zhang, Karl Pertsch et al.

We propose BOSS, an approach that automatically learns to solve new long-horizon, complex, and meaningful tasks by growing a learned skill library with minimal supervision. Prior work in reinforcement learning require expert supervision, in the form of demonstrations or rich reward functions, to learn long-horizon tasks. Instead, our approach BOSS (BOotStrapping your own Skills) learns to accomplish new tasks by performing "skill bootstrapping," where an agent with a set of primitive skills interacts with the environment to practice new skills without receiving reward feedback for tasks outside of the initial skill set. This bootstrapping phase is guided by large language models (LLMs) that inform the agent of meaningful skills to chain together. Through this process, BOSS builds a wide range of complex and useful behaviors from a basic set of primitive skills. We demonstrate through experiments in realistic household environments that agents trained with our LLM-guided bootstrapping procedure outperform those trained with naive bootstrapping as well as prior unsupervised skill acquisition methods on zero-shot execution of unseen, long-horizon tasks in new environments. Website at clvrai.com/boss.

Chandler Smith, Marwa Abdulhai, Manfred Diaz et al.

Large Language Model (LLM) agents have demonstrated impressive capabilities for social interaction and are increasingly being deployed in situations where they might engage with both human and artificial agents. These interactions represent a critical frontier for LLM-based agents, yet existing evaluation methods fail to measure how well these capabilities generalize to novel social situations. In this paper, we introduce a method for evaluating the ability of LLM-based agents to cooperate in zero-shot, mixed-motive environments using Concordia, a natural language multi-agent simulation environment. Our method measures general cooperative intelligence by testing an agent's ability to identify and exploit opportunities for mutual gain across diverse partners and contexts. We present empirical results from the NeurIPS 2024 Concordia Contest, where agents were evaluated on their ability to achieve mutual gains across a suite of diverse scenarios ranging from negotiation to collective action problems. Our findings reveal significant gaps between current agent capabilities and the robust generalization required for reliable cooperation, particularly in scenarios demanding persuasion and norm enforcement.

Jeongeun Park, Seungwon Lim, Joonhyung Lee et al.

In this paper, we focus on inferring whether the given user command is clear, ambiguous, or infeasible in the context of interactive robotic agents utilizing large language models (LLMs). To tackle this problem, we first present an uncertainty estimation method for LLMs to classify whether the command is certain (i.e., clear) or not (i.e., ambiguous or infeasible). Once the command is classified as uncertain, we further distinguish it between ambiguous or infeasible commands leveraging LLMs with situational aware context in a zero-shot manner. For ambiguous commands, we disambiguate the command by interacting with users via question generation with LLMs. We believe that proper recognition of the given commands could lead to a decrease in malfunction and undesired actions of the robot, enhancing the reliability of interactive robot agents. We present a dataset for robotic situational awareness, consisting pair of high-level commands, scene descriptions, and labels of command type (i.e., clear, ambiguous, or infeasible). We validate the proposed method on the collected dataset, pick-and-place tabletop simulation. Finally, we demonstrate the proposed approach in real-world human-robot interaction experiments, i.e., handover scenarios.

Joel Z. Leibo, Alexander Sasha Vezhnevets, Manfred Diaz et al.

We propose a society-first theory of normative appropriateness where individuals, modeled as pre-trained actors with cognitive architectures analogous to Large Language Models (LLMs), generate behavior via predictive pattern completion. Our theory posits that individuals act by completing distributed symbolic patterns based on context, answering questions such as "What does a person such as I do in a situation such as this?". This sense-making mechanism provides a parsimonious account of the key features of human norms: their context-dependence, arbitrariness, automaticity, dynamism, and their support from social sanctioning. It challenges rational-choice theories of social norms by accounting for their key features without needing to exogenously posit scalar rewards or preference relations. By distinguishing between explicit norms, which we associate with in-context adaptation, and implicit norms, which we associate with long-term memory, the theory reconceptualizes several foundational ideas in cognitive science. In particular, it gives an alternative account to the data traditionally seen as supporting dual-process models, and it flips the role of rationality, allowing us to construe it as adherence to culturally-contingent justification standards.

Miyoung Ko, Ingyu Seong, Hwaran Lee et al.

With the growing importance of detecting misinformation, many studies have focused on verifying factual claims by retrieving evidence. However, canonical fact verification tasks do not apply to catching subtle differences in factually consistent claims, which might still bias the readers, especially on contentious political or economic issues. Our underlying assumption is that among the trusted sources, one's argument is not necessarily more true than the other, requiring comparison rather than verification. In this study, we propose ClaimDiff, a novel dataset that primarily focuses on comparing the nuance between claim pairs. In ClaimDiff, we provide 2,941 annotated claim pairs from 268 news articles. We observe that while humans are capable of detecting the nuances between claims, strong baselines struggle to detect them, showing over a 19% absolute gap with the humans. We hope this initial study could help readers to gain an unbiased grasp of contentious issues through machine-aided comparison.

Young-Ho Kim, Sungdong Kim, Minsuk Chang et al.

Current natural language interaction for self-tracking tools largely depends on bespoke implementation optimized for a specific tracking theme and data format, which is neither generalizable nor scalable to a tremendous design space of self-tracking. However, training machine learning models in the context of self-tracking is challenging due to the wide variety of tracking topics and data formats. In this paper, we propose a novel NLP task for self-tracking that extracts close- and open-ended information from a retrospective activity log described as a plain text, and a domain-agnostic, GPT-3-based NLU framework that performs this task. The framework augments the prompt using synthetic samples to transform the task into 10-shot learning, to address a cold-start problem in bootstrapping a new tracking topic. Our preliminary evaluation suggests that our approach significantly outperforms the baseline QA models. Going further, we discuss future application domains toward which the NLP and HCI researchers can collaborate.

Tae Soo Kim, Arghya Sarkar, Yoonjoo Lee et al.

Large language models (LLMs) can enhance writing by automating or supporting specific tasks in writers' workflows (e.g., paraphrasing, creating analogies). Leveraging this capability, a collection of interfaces have been developed that provide LLM-powered tools for specific writing tasks. However, these interfaces provide limited support for writers to create personal tools for their own unique tasks, and may not comprehensively fulfill a writer's needs -- requiring them to continuously switch between interfaces during writing. In this work, we envision LMCanvas, an interface that enables writers to create their own LLM-powered writing tools and arrange their personal writing environment by interacting with "blocks" in a canvas. In this interface, users can create text blocks to encapsulate writing and LLM prompts, model blocks for model parameter configurations, and connect these to create pipeline blocks that output generations. In this workshop paper, we discuss the design for LMCanvas and our plans to develop this concept.

Soohyun Lee, Minsuk Chang, Seokhyeon Park et al.

Recent advancements in vision models have greatly improved their ability to handle complex chart understanding tasks, like chart captioning and question answering. However, it remains challenging to assess how these models process charts. Existing benchmarks only roughly evaluate model performance without evaluating the underlying mechanisms, such as how models extract image embeddings. This limits our understanding of the model's ability to perceive fundamental graphical components. To address this, we introduce a novel evaluation framework to assess the graphical perception of image embedding models. For chart comprehension, we examine two main aspects of channel effectiveness: accuracy and discriminability of various visual channels. Channel accuracy is assessed through the linearity of embeddings, measuring how well the perceived magnitude aligns with the size of the stimulus. Discriminability is evaluated based on the distances between embeddings, indicating their distinctness. Our experiments with the CLIP model show that it perceives channel accuracy differently from humans and shows unique discriminability in channels like length, tilt, and curvature. We aim to develop this work into a broader benchmark for reliable visual encoders, enhancing models for precise chart comprehension and human-like perception in future applications.

Minsuk Kahng, Ian Tenney, Mahima Pushkarna et al. · deepmind

Automatic side-by-side evaluation has emerged as a promising approach to evaluating the quality of responses from large language models (LLMs). However, analyzing the results from this evaluation approach raises scalability and interpretability challenges. In this paper, we present LLM Comparator, a novel visual analytics tool for interactively analyzing results from automatic side-by-side evaluation. The tool supports interactive workflows for users to understand when and why a model performs better or worse than a baseline model, and how the responses from two models are qualitatively different. We iteratively designed and developed the tool by closely working with researchers and engineers at a large technology company. This paper details the user challenges we identified, the design and development of the tool, and an observational study with participants who regularly evaluate their models.

Nitesh Goyal, Minsuk Chang, Michael Terry

Our ability to build autonomous agents that leverage Generative AI continues to increase by the day. As builders and users of such agents it is unclear what parameters we need to align on before the agents start performing tasks on our behalf. To discover these parameters, we ran a qualitative empirical research study about designing agents that can negotiate during a fictional yet relatable task of selling a camera online. We found that for an agent to perform the task successfully, humans/users and agents need to align over 6 dimensions: 1) Knowledge Schema Alignment 2) Autonomy and Agency Alignment 3) Operational Alignment and Training 4) Reputational Heuristics Alignment 5) Ethics Alignment and 6) Human Engagement Alignment. These empirical findings expand previous work related to process and specification alignment and the need for values and safety in Human-AI interactions. Subsequently we discuss three design directions for designers who are imagining a world filled with Human-Agent collaborations.

JiWoo Kim, Minsuk Chang, JinYeong Bak

Traditional text-based human-AI interactions often adhere to a strict turn-taking approach. In this research, we propose a novel approach that incorporates overlapping messages, mirroring natural human conversations. Through a formative study, we observed that even in text-based contexts, users instinctively engage in overlapping behaviors like "A: Today I went to-" "B: yeah." To capitalize on these insights, we developed OverlapBot, a prototype chatbot where both AI and users can initiate overlapping. Our user study revealed that OverlapBot was perceived as more communicative and immersive than traditional turn-taking chatbot, fostering faster and more natural interactions. Our findings contribute to the understanding of design space for overlapping interactions. We also provide recommendations for implementing overlap-capable AI interactions to enhance the fluidity and engagement of text-based conversations.

Alexander Sasha Vezhnevets, Jayd Matyas, Logan Cross et al.

Generative AI can be used in multi-actor environments with purposes ranging from social science modeling to interactive narrative and AI evaluation. Supporting this diversity of use cases -- which we classify as Simulationist, Dramatist, and Evaluationist -- demands a flexible scenario definition framework. We argue here that a good approach is to take inspiration from tabletop role-playing games (TTRPGs), where a Game Master (GM) is responsible for the environment and generates all parts of the story not directly determined by the voluntary actions of player characters. We argue that the Entity-Component architectural pattern is useful here. In such a system, the GM is not a hardcoded computer game but is itself a configurable entity, composed of components just like any other actor. By design, the approach allows for a separation between the underlying implementation details handled by an engineer, the creation of reusable components, and their composition and configuration managed by a designer who constructs entities from the components. This separation of concerns is instrumental for achieving rapid iteration, maintaining modularity, and ultimately to ensure scalability. We describe the ongoing evolution of the Concordia library in terms of this philosophy, demonstrating how it allows users to effectively configure scenarios that align with their specific goals.

Minsuk Chang, Yao Wang, Huichen Will Wang et al. · uw

Accounting for individual differences can improve the effectiveness of visualization design. While the role of visual attention in visualization interpretation is well recognized, existing work often overlooks how this behavior varies based on visual literacy levels. Based on data from a 235-participant user study covering three visualization tests (mini-VLAT, CALVI, and SGL), we show that distinct attention patterns in visual data exploration can correlate with participants' literacy levels: While experts (high-scorers) generally show a strong attentional focus, novices (low-scorers) focus less and explore more. We then propose two computational models leveraging these insights: Lit2Sal -- a novel visual saliency model that predicts observer attention given their visualization literacy level, and Sal2Lit -- a model to predict visual literacy from human visual attention data. Our quantitative and qualitative evaluation demonstrates that Lit2Sal outperforms state-of-the-art saliency models with literacy-aware considerations. Sal2Lit predicts literacy with 86% accuracy using a single attention map, providing a time-efficient supplement to literacy assessment that only takes less than a minute. Taken together, our unique approach to consider individual differences in salience models and visual attention in literacy assessments paves the way for new directions in personalized visual data communication to enhance understanding.

Katherine M. Collins, Najoung Kim, Yonatan Bitton et al.

Human feedback plays a critical role in learning and refining reward models for text-to-image generation, but the optimal form the feedback should take for learning an accurate reward function has not been conclusively established. This paper investigates the effectiveness of fine-grained feedback which captures nuanced distinctions in image quality and prompt-alignment, compared to traditional coarse-grained feedback (for example, thumbs up/down or ranking between a set of options). While fine-grained feedback holds promise, particularly for systems catering to diverse societal preferences, we show that demonstrating its superiority to coarse-grained feedback is not automatic. Through experiments on real and synthetic preference data, we surface the complexities of building effective models due to the interplay of model choice, feedback type, and the alignment between human judgment and computational interpretation. We identify key challenges in eliciting and utilizing fine-grained feedback, prompting a reassessment of its assumed benefits and practicality. Our findings -- e.g., that fine-grained feedback can lead to worse models for a fixed budget, in some settings; however, in controlled settings with known attributes, fine grained rewards can indeed be more helpful -- call for careful consideration of feedback attributes and potentially beckon novel modeling approaches to appropriately unlock the potential value of fine-grained feedback in-the-wild.

Minsuk Chang, Seokhyeon Park, Hyeon Jeon et al.

In image classification, a significant problem arises from bias in the datasets. When it contains only specific types of images, the classifier begins to rely on shortcuts - simplistic and erroneous rules for decision-making. This leads to high performance on the training dataset but inferior results on new, varied images, as the classifier's generalization capability is reduced. For example, if the images labeled as mustache consist solely of male figures, the model may inadvertently learn to classify images by gender rather than the presence of a mustache. One approach to mitigate such biases is to direct the model's attention toward the target object's location, usually marked using bounding boxes or polygons for annotation. However, collecting such annotations requires substantial time and human effort. Therefore, we propose a novel patch-labeling method that integrates AI assistance with crowdsourcing to capture human attention from images, which can be a viable solution for mitigating bias. Our method consists of two steps. First, we extract the approximate location of a target using a pre-trained saliency detection model supplemented by human verification for accuracy. Then, we determine the human-attentive area in the image by iteratively dividing the image into smaller patches and employing crowdsourcing to ascertain whether each patch can be classified as the target object. We demonstrated the effectiveness of our method in mitigating bias through improved classification accuracy and the refined focus of the model. Also, crowdsourced experiments validate that our method collects human annotation up to 3.4 times faster than annotating object locations with polygons, significantly reducing the need for human resources. We conclude the paper by discussing the advantages of our method in a crowdsourcing context, mainly focusing on aspects of human errors and accessibility.

Boseop Kim, HyoungSeok Kim, Sang-Woo Lee et al.

GPT-3 shows remarkable in-context learning ability of large-scale language models (LMs) trained on hundreds of billion scale data. Here we address some remaining issues less reported by the GPT-3 paper, such as a non-English LM, the performances of different sized models, and the effect of recently introduced prompt optimization on in-context learning. To achieve this, we introduce HyperCLOVA, a Korean variant of 82B GPT-3 trained on a Korean-centric corpus of 560B tokens. Enhanced by our Korean-specific tokenization, HyperCLOVA with our training configuration shows state-of-the-art in-context zero-shot and few-shot learning performances on various downstream tasks in Korean. Also, we show the performance benefits of prompt-based learning and demonstrate how it can be integrated into the prompt engineering pipeline. Then we discuss the possibility of materializing the No Code AI paradigm by providing AI prototyping capabilities to non-experts of ML by introducing HyperCLOVA studio, an interactive prompt engineering interface. Lastly, we demonstrate the potential of our methods with three successful in-house applications.

Sungdong Kim, Minsuk Chang, Sang-Woo Lee

We propose NeuralWOZ, a novel dialogue collection framework that uses model-based dialogue simulation. NeuralWOZ has two pipelined models, Collector and Labeler. Collector generates dialogues from (1) user's goal instructions, which are the user context and task constraints in natural language, and (2) system's API call results, which is a list of possible query responses for user requests from the given knowledge base. Labeler annotates the generated dialogue by formulating the annotation as a multiple-choice problem, in which the candidate labels are extracted from goal instructions and API call results. We demonstrate the effectiveness of the proposed method in the zero-shot domain transfer learning for dialogue state tracking. In the evaluation, the synthetic dialogue corpus generated from NeuralWOZ achieves a new state-of-the-art with improvements of 4.4% point joint goal accuracy on average across domains, and improvements of 5.7% point of zero-shot coverage against the MultiWOZ 2.1 dataset.