Bryan Wang

Stephen Brade, Bryan Wang, Mauricio Sousa et al. · utoronto

Text-to-image generative models have demonstrated remarkable capabilities in generating high-quality images based on textual prompts. However, crafting prompts that accurately capture the user's creative intent remains challenging. It often involves laborious trial-and-error procedures to ensure that the model interprets the prompts in alignment with the user's intention. To address the challenges, we present Promptify, an interactive system that supports prompt exploration and refinement for text-to-image generative models. Promptify utilizes a suggestion engine powered by large language models to help users quickly explore and craft diverse prompts. Our interface allows users to organize the generated images flexibly, and based on their preferences, Promptify suggests potential changes to the original prompt. This feedback loop enables users to iteratively refine their prompts and enhance desired features while avoiding unwanted ones. Our user study shows that Promptify effectively facilitates the text-to-image workflow and outperforms an existing baseline tool widely used for text-to-image generation.

Bryan Wang, Gang Li, Yang Li

Conversational agents show the promise to allow users to interact with mobile devices using language. However, to perform diverse UI tasks with natural language, developers typically need to create separate datasets and models for each specific task, which is expensive and effort-consuming. Recently, pre-trained large language models (LLMs) have been shown capable of generalizing to various downstream tasks when prompted with a handful of examples from the target task. This paper investigates the feasibility of enabling versatile conversational interactions with mobile UIs using a single LLM. We designed prompting techniques to adapt an LLM to mobile UIs. We experimented with four important modeling tasks that address various scenarios in conversational interaction. Our method achieved competitive performance on these challenging tasks without requiring dedicated datasets and training, offering a lightweight and generalizable approach to enable language-based mobile interaction.

Zhecheng Wang, Jiaju Ma, Eitan Grinspun et al.

Scriptwriting has traditionally been text-centric, a modality that only partially conveys the produced audiovisual experience. A formative study with professional writers informed us that connecting textual and audiovisual modalities can aid ideation and iteration, especially for writing dialogues. In this work, we present Script2Screen, an AI-assisted tool that integrates scriptwriting with audiovisual scene creation in a unified, synchronized workflow. Focusing on dialogues in scripts, Script2Screen generates expressive scenes with emotional speeches and animated characters through a novel text-to-audiovisual-scene pipeline. The user interface provides fine-grained controls, allowing writers to fine-tune audiovisual elements such as character gestures, speech emotions, and camera angles. A user study with both novice and professional writers from various domains demonstrated that Script2Screen's interactive audiovisual generation enhances the scriptwriting process, facilitating iterative refinement while complementing, rather than replacing, their creative efforts.

Yilun Zhu, Yuan Zhuang, Nikhita Vedula et al.

Many applications of LLM-based text regression require predicting a full conditional distribution rather than a single point value. We study distributional regression under empirical-quantile supervision, where each input is paired with multiple observed quantile outcomes, and the target distribution is represented by a dense grid of quantiles. We address two key limitations of current approaches: the lack of local grounding for distribution estimates, and the reliance on shared representations that create an indirect bottleneck between inputs and quantile outputs. In this paper, we introduce Quantile Token Regression, which, to our knowledge, is the first work to insert dedicated quantile tokens into the input sequence, enabling direct input-output pathways for each quantile through self-attention. We further augment these quantile tokens with retrieval, incorporating semantically similar neighbor instances and their empirical distributions to ground predictions with local evidence from similar instances. We also provide the first theoretical analysis of loss functions for quantile regression, clarifying which distributional objectives each optimizes. Experiments on the Inside Airbnb and StackSample benchmark datasets with LLMs ranging from 1.7B to 14B parameters show that quantile tokens with neighbors consistently outperform baselines (~4 points lower MAPE and 2x narrower prediction intervals), with especially large gains on smaller and more challenging datasets where quantile tokens produce substantially sharper and more accurate distributions.

Minghe Zhang, Chaosheng Dong, Jinmiao Fu et al.

In this paper, we introduce AdaSelection, an adaptive sub-sampling method to identify the most informative sub-samples within each minibatch to speed up the training of large-scale deep learning models without sacrificing model performance. Our method is able to flexibly combines an arbitrary number of baseline sub-sampling methods incorporating the method-level importance and intra-method sample-level importance at each iteration. The standard practice of ad-hoc sampling often leads to continuous training with vast amounts of data from production environments. To improve the selection of data instances during forward and backward passes, we propose recording a constant amount of information per instance from these passes. We demonstrate the effectiveness of our method by testing it across various types of inputs and tasks, including the classification tasks on both image and language datasets, as well as regression tasks. Compared with industry-standard baselines, AdaSelection consistently displays superior performance.

Moyan Li, Jinmiao Fu, Shaoyuan Xu et al.

On shopping websites, product images of low quality negatively affect customer experience. Although there are plenty of work in detecting images with different defects, few efforts have been dedicated to correct those defects at scale. A major challenge is that there are thousands of product types and each has specific defects, therefore building defect specific models is unscalable. In this paper, we propose a unified Image-to-Image (I2I) translation model to correct multiple defects across different product types. Our model leverages an attention mechanism to hierarchically incorporate high-level defect groups and specific defect types to guide the network to focus on defect-related image regions. Evaluated on eight public datasets, our model reduces the Frechet Inception Distance (FID) by 24.6% in average compared with MoNCE, the state-of-the-art I2I method. Unlike public data, another practical challenge on shopping websites is that some paired images are of low quality. Therefore we design our model to be semi-paired by combining the L1 loss of paired data with the cycle loss of unpaired data. Tested on a shopping website dataset to correct three image defects, our model reduces (FID) by 63.2% in average compared with WS-I2I, the state-of-the art semi-paired I2I method.

Simon Valentin, Jinmiao Fu, Gianluca Detommaso et al.

Large language models (LLMs) can be prone to hallucinations - generating unreliable outputs that are unfaithful to their inputs, external facts or internally inconsistent. In this work, we address several challenges for post-hoc hallucination detection in production settings. Our pipeline for hallucination detection entails: first, producing a confidence score representing the likelihood that a generated answer is a hallucination; second, calibrating the score conditional on attributes of the inputs and candidate response; finally, performing detection by thresholding the calibrated score. We benchmark a variety of state-of-the-art scoring methods on different datasets, encompassing question answering, fact checking, and summarization tasks. We employ diverse LLMs to ensure a comprehensive assessment of performance. We show that calibrating individual scoring methods is critical for ensuring risk-aware downstream decision making. Based on findings that no individual score performs best in all situations, we propose a multi-scoring framework, which combines different scores and achieves top performance across all datasets. We further introduce cost-effective multi-scoring, which can match or even outperform more expensive detection methods, while significantly reducing computational overhead.

Tao Li, Gang Li, Zhiwei Deng et al.

Large language models (LLMs) have shown increasing capacity at planning and executing a high-level goal in a live computer environment (e.g. MiniWoB++). To perform a task, recent works often require a model to learn from trace examples of the task via either supervised learning or few/many-shot prompting. Without these trace examples, it remains a challenge how an agent can autonomously learn and improve its control on a computer, which limits the ability of an agent to perform a new task. We approach this problem with a zero-shot agent that requires no given expert traces. Our agent plans for executable actions on a partially observed environment, and iteratively progresses a task by identifying and learning from its mistakes via self-reflection and structured thought management. On the easy tasks of MiniWoB++, we show that our zero-shot agent often outperforms recent SoTAs, with more efficient reasoning. For tasks with more complexity, our reflective agent performs on par with prior best models, even though previous works had the advantages of accessing expert traces or additional screen information.

Chien-Chih Wang, Shaoyuan Xu, Jinmiao Fu et al.

Semi-supervised learning (SSL) has become a crucial approach in deep learning as a way to address the challenge of limited labeled data. The success of deep neural networks heavily relies on the availability of large-scale high-quality labeled data. However, the process of data labeling is time-consuming and unscalable, leading to shortages in labeled data. SSL aims to tackle this problem by leveraging additional unlabeled data in the training process. One of the popular SSL algorithms, FixMatch, trains identical weight-sharing teacher and student networks simultaneously using a siamese neural network (SNN). However, it is prone to performance degradation when the pseudo labels are heavily noisy in the early training stage. We present KD-FixMatch, a novel SSL algorithm that addresses the limitations of FixMatch by incorporating knowledge distillation. The algorithm utilizes a combination of sequential and simultaneous training of SNNs to enhance performance and reduce performance degradation. Firstly, an outer SNN is trained using labeled and unlabeled data. After that, the network of the well-trained outer SNN generates pseudo labels for the unlabeled data, from which a subset of unlabeled data with trusted pseudo labels is then carefully created through high-confidence sampling and deep embedding clustering. Finally, an inner SNN is trained with the labeled data, the unlabeled data, and the subset of unlabeled data with trusted pseudo labels. Experiments on four public data sets demonstrate that KD-FixMatch outperforms FixMatch in all cases. Our results indicate that KD-FixMatch has a better training starting point that leads to improved model performance compared to FixMatch.

Lingkai Kong, Molei Tao, Yang Liu et al.

Flow-based Generative Models (FGMs) effectively transform noise into complex data distributions. Incorporating Optimal Transport (OT) to couple noise and data during FGM training has been shown to improve the straightness of flow trajectories, enabling more effective inference. However, existing OT-based methods estimate the OT plan using (mini-)batches of sampled noise and data points, which limits their scalability to large and high-dimensional datasets in FGMs. This paper introduces AlignFlow, a novel approach that leverages Semi-Discrete Optimal Transport (SDOT) to enhance the training of FGMs by establishing an explicit, optimal alignment between noise distribution and data points with guaranteed convergence. SDOT computes a transport map by partitioning the noise space into Laguerre cells, each mapped to a corresponding data point. During FGM training, i.i.d. noise samples are paired with data points via the SDOT map. AlignFlow scales well to large datasets and model architectures with negligible computational overhead. Experimental results show that AlignFlow improves the performance of a wide range of state-of-the-art FGM algorithms and can be integrated as a plug-and-play component. Code is available at: https://github.com/konglk1203/AlignFlow.

Bryan Wang, Gang Li, Xin Zhou et al.

Mobile User Interface Summarization generates succinct language descriptions of mobile screens for conveying important contents and functionalities of the screen, which can be useful for many language-based application scenarios. We present Screen2Words, a novel screen summarization approach that automatically encapsulates essential information of a UI screen into a coherent language phrase. Summarizing mobile screens requires a holistic understanding of the multi-modal data of mobile UIs, including text, image, structures as well as UI semantics, motivating our multi-modal learning approach. We collected and analyzed a large-scale screen summarization dataset annotated by human workers. Our dataset contains more than 112k language summarization across $\sim$22k unique UI screens. We then experimented with a set of deep models with different configurations. Our evaluation of these models with both automatic accuracy metrics and human rating shows that our approach can generate high-quality summaries for mobile screens. We demonstrate potential use cases of Screen2Words and open-source our dataset and model to lay the foundations for further bridging language and user interfaces.

Yu-Hua Chen, Bryan Wang, Yi-Hsuan Yang

We present in this paper PerformacnceNet, a neural network model we proposed recently to achieve score-to-audio music generation. The model learns to convert a music piece from the symbolic domain to the audio domain, assigning performance-level attributes such as changes in velocity automatically to the music and then synthesizing the audio. The model is therefore not just a neural audio synthesizer, but an AI performer that learns to interpret a musical score in its own way. The code and sample outputs of the model can be found online at https://github.com/bwang514/PerformanceNet.

Bryan Wang, Yi-Hsuan Yang

Music creation is typically composed of two parts: composing the musical score, and then performing the score with instruments to make sounds. While recent work has made much progress in automatic music generation in the symbolic domain, few attempts have been made to build an AI model that can render realistic music audio from musical scores. Directly synthesizing audio with sound sample libraries often leads to mechanical and deadpan results, since musical scores do not contain performance-level information, such as subtle changes in timing and dynamics. Moreover, while the task may sound like a text-to-speech synthesis problem, there are fundamental differences since music audio has rich polyphonic sounds. To build such an AI performer, we propose in this paper a deep convolutional model that learns in an end-to-end manner the score-to-audio mapping between a symbolic representation of music called the piano rolls and an audio representation of music called the spectrograms. The model consists of two subnets: the ContourNet, which uses a U-Net structure to learn the correspondence between piano rolls and spectrograms and to give an initial result; and the TextureNet, which further uses a multi-band residual network to refine the result by adding the spectral texture of overtones and timbre. We train the model to generate music clips of the violin, cello, and flute, with a dataset of moderate size. We also present the result of a user study that shows our model achieves higher mean opinion score (MOS) in naturalness and emotional expressivity than a WaveNet-based model and two commercial sound libraries. We open our source code at https://github.com/bwang514/PerformanceNet

Bryan Wang, Yuliang Li, Zhaoyang Lv et al.

Video creation has become increasingly popular, yet the expertise and effort required for editing often pose barriers to beginners. In this paper, we explore the integration of large language models (LLMs) into the video editing workflow to reduce these barriers. Our design vision is embodied in LAVE, a novel system that provides LLM-powered agent assistance and language-augmented editing features. LAVE automatically generates language descriptions for the user's footage, serving as the foundation for enabling the LLM to process videos and assist in editing tasks. When the user provides editing objectives, the agent plans and executes relevant actions to fulfill them. Moreover, LAVE allows users to edit videos through either the agent or direct UI manipulation, providing flexibility and enabling manual refinement of agent actions. Our user study, which included eight participants ranging from novices to proficient editors, demonstrated LAVE's effectiveness. The results also shed light on user perceptions of the proposed LLM-assisted editing paradigm and its impact on users' creativity and sense of co-creation. Based on these findings, we propose design implications to inform the future development of agent-assisted content editing.

Catherine Yeh, Anh Truong, Mira Dontcheva et al.

Video storytelling is often constrained by available material, limiting creative expression and leaving undesired narrative gaps. Generative video offers a new way to address these limitations by augmenting captured media with tailored visuals. To explore this potential, we interviewed eight video creators to identify opportunities and challenges in integrating generative video into their workflows. Building on these insights and established filmmaking principles, we developed Vidmento, a tool for authoring hybrid video stories that combine captured and generated media through context-aware expansion. Vidmento surfaces opportunities for story development, generates clips that blend stylistically and narratively with surrounding media, and provides controls for refinement. In a study with 12 creators, Vidmento supported narrative development and exploration by systematically expanding initial materials with generative media, enabling expressive video storytelling aligned with creative intent. We highlight how creators bridge story gaps with generative content and where they find this blending capability most valuable.

Yanhui Guo, Shaoyuan Xu, Jinmiao Fu et al.

This paper introduces \textbf{Q-tuning}, a novel approach for continual prompt tuning that enables the lifelong learning of a pre-trained language model. When learning a new task, Q-tuning trains a task-specific prompt by adding it to a prompt queue consisting of the prompts from older tasks. To better transfer the knowledge of old tasks, we design an adaptive knowledge aggregation technique that reweighs previous prompts in the queue with a learnable low-rank matrix. Once the prompt queue reaches its maximum capacity, we leverage a PCA-based eviction rule to reduce the queue's size, allowing the newly trained prompt to be added while preserving the primary knowledge of old tasks. In order to mitigate the accumulation of information loss caused by the eviction, we additionally propose a globally shared prefix prompt and a memory retention regularization based on information theory. Extensive experiments demonstrate that our approach outperforms the state-of-the-art methods substantially on continual prompt tuning benchmarks. Moreover, our approach enables lifelong learning on linearly growing task sequences while requiring constant complexity for training and inference.

Stephen Brade, Bryan Wang, Mauricio Sousa et al.

Synthesizers are powerful tools that allow musicians to create dynamic and original sounds. Existing commercial interfaces for synthesizers typically require musicians to interact with complex low-level parameters or to manage large libraries of premade sounds. To address these challenges, we implement SynthScribe -- a fullstack system that uses multimodal deep learning to let users express their intentions at a much higher level. We implement features which address a number of difficulties, namely 1) searching through existing sounds, 2) creating completely new sounds, 3) making meaningful modifications to a given sound. This is achieved with three main features: a multimodal search engine for a large library of synthesizer sounds; a user centered genetic algorithm by which completely new sounds can be created and selected given the users preferences; a sound editing support feature which highlights and gives examples for key control parameters with respect to a text or audio based query. The results of our user studies show SynthScribe is capable of reliably retrieving and modifying sounds while also affording the ability to create completely new sounds that expand a musicians creative horizon.

Hengkang Wang, Yang Liu, Huidong Liu et al.

Video restoration (VR) aims to recover high-quality videos from degraded ones. Although recent zero-shot VR methods using pre-trained diffusion models (DMs) show good promise, they suffer from approximation errors during reverse diffusion and insufficient temporal consistency. Moreover, dealing with 3D video data, VR is inherently computationally intensive. In this paper, we advocate viewing the reverse process in DMs as a function and present a novel Maximum a Posterior (MAP) framework that directly parameterizes video frames in the seed space of DMs, eliminating approximation errors. We also introduce strategies to promote bilevel temporal consistency: semantic consistency by leveraging clustering structures in the seed space, and pixel-level consistency by progressive warping with optical flow refinements. Extensive experiments on multiple virtual reality tasks demonstrate superior visual quality and temporal consistency achieved by our method compared to the state-of-the-art.

Anirban Saha Anik, Xiaoying Song, Elliott Wang et al.

Large language models (LLMs) incorporated with Retrieval-Augmented Generation (RAG) have demonstrated powerful capabilities in generating counterspeech against misinformation. However, current studies rely on limited evidence and offer less control over final outputs. To address these challenges, we propose a Multi-agent Retrieval-Augmented Framework to generate counterspeech against health misinformation, incorporating multiple LLMs to optimize knowledge retrieval, evidence enhancement, and response refinement. Our approach integrates both static and dynamic evidence, ensuring that the generated counterspeech is relevant, well-grounded, and up-to-date. Our method outperforms baseline approaches in politeness, relevance, informativeness, and factual accuracy, demonstrating its effectiveness in generating high-quality counterspeech. To further validate our approach, we conduct ablation studies to verify the necessity of each component in our framework. Furthermore, cross evaluations show that our system generalizes well across diverse health misinformation topics and datasets. And human evaluations reveal that refinement significantly enhances counterspeech quality and obtains human preference.

Xinhai Hou, Shaoyuan Xu, Manan Biyani et al.

Agentic vision-language models are increasingly trained to "think with images" by calling image operations. However, we show that high final-answer accuracy often hides unfaithful visual reasoning: models may invoke tools on irrelevant regions or ignore tool outputs entirely, yet still guess the correct answer. In this work, we first propose a faithfulness evaluation protocol that measures whether intermediate visual tool outputs (e.g., crops) actually contain the queried evidence. This reveals that recent visual agents achieve high final-answer accuracy but exhibit low rates of faithful tool-use on visual search benchmarks. We then introduce CodeV, a code-based visual agent trained with Tool-Aware Policy Optimization (TAPO). TAPO is a process-level RL framework that augments GRPO with dense rewards defined directly on visual tool inputs and outputs, rather than on chain-of-thought tokens, making supervision easier to verify and less susceptible to reward hacking. CodeV represents visual tools as executable Python code, and TAPO assigns step-wise rewards based solely on the question and tool output, encouraging both necessary and evidence-consistent tool use. In a two-stage SFT+RL pipeline, CodeV achieves competitive or superior accuracy while substantially increasing faithful tool-use rates on related visual search benchmarks. Beyond visual search, CodeV attains strong performance on a range of multimodal reasoning and math benchmarks, suggesting that explicitly supervising intermediate tool behavior is crucial for building trustworthy, agentic visual reasoning systems.

Aditya Gunturu, Ben Pearman, Keiichi Ihara et al.

We introduce MapStory, an LLM-powered animation prototyping tool that generates editable map animation sequences directly from natural language text by leveraging a dual-agent LLM architecture. Given a user written script, MapStory automatically produces a scene breakdown, which decomposes the text into key map animation primitives such as camera movements, visual highlights, and animated elements. Our system includes a researcher agent that accurately queries geospatial information by leveraging an LLM with web search, enabling automatic extraction of relevant regions, paths, and coordinates while allowing users to edit and query for changes or additional information to refine the results. Additionally, users can fine-tune parameters of these primitive blocks through an interactive timeline editor. We detail the system's design and architecture, informed by formative interviews with professional animators and by an analysis of 200 existing map animation videos. Our evaluation, which includes expert interviews (N=5) and a usability study (N=12), demonstrates that MapStory enables users to create map animations with ease, facilitates faster iteration, encourages creative exploration, and lowers barriers to creating map-centric stories.

Binh M. Le, Shaoyuan Xu, Jinmiao Fu et al.

In Visual Document Understanding (VDU) tasks, fine-tuning a pre-trained Vision-Language Model (VLM) with new datasets often falls short in optimizing the vision encoder to identify query-specific regions in text-rich document images. Existing methods that directly inject queries into model layers by modifying the network architecture often struggle to adapt to new datasets with limited annotations. To address this, we introduce QID, a novel, streamlined, architecture-preserving approach that integrates query embeddings into the vision encoder, leading to notable performance gains, particularly in data-scarce fine-tuning scenarios. Specifically, our approach introduces a dual-module framework: a query-aware module that generates a unique query vector to precisely guide the model's focus, as well as a query-agnostic module that captures the positional relationships among tokens, ensuring robust spatial understanding. Notably, both modules operate independently of the vision attention blocks, facilitating targeted learning of query embeddings and enhancing visual semantic identification. Experiments with OCR-free VLMs across multiple datasets demonstrate significant performance improvements using our method, especially in handling text-rich documents in data-scarce environments.

Huidong Liu, Shaoyuan Xu, Jinmiao Fu et al.

Modern Web systems such as social media and e-commerce contain rich contents expressed in images and text. Leveraging information from multi-modalities can improve the performance of machine learning tasks such as classification and recommendation. In this paper, we propose the Cross-Modality Attention Contrastive Language-Image Pre-training (CMA-CLIP), a new framework which unifies two types of cross-modality attentions, sequence-wise attention and modality-wise attention, to effectively fuse information from image and text pairs. The sequence-wise attention enables the framework to capture the fine-grained relationship between image patches and text tokens, while the modality-wise attention weighs each modality by its relevance to the downstream tasks. In addition, by adding task specific modality-wise attentions and multilayer perceptrons, our proposed framework is capable of performing multi-task classification with multi-modalities. We conduct experiments on a Major Retail Website Product Attribute (MRWPA) dataset and two public datasets, Food101 and Fashion-Gen. The results show that CMA-CLIP outperforms the pre-trained and fine-tuned CLIP by an average of 11.9% in recall at the same level of precision on the MRWPA dataset for multi-task classification. It also surpasses the state-of-the-art method on Fashion-Gen Dataset by 5.5% in accuracy and achieves competitive performance on Food101 Dataset. Through detailed ablation studies, we further demonstrate the effectiveness of both cross-modality attention modules and our method's robustness against noise in image and text inputs, which is a common challenge in practice.

Zhuoyue Lyu, Jiannan Li, Bryan Wang

Artificial Intelligence (AI), especially Neural Networks (NNs), has become increasingly popular. However, people usually treat AI as a tool, focusing on improving outcome, accuracy, and performance while paying less attention to the representation of AI itself. We present AIive, an interactive visualization of AI in Virtual Reality (VR) that brings AI "alive". AIive enables users to manipulate the parameters of NNs with virtual hands and provides auditory feedback for the real-time values of loss, accuracy, and hyperparameters. Thus, AIive contributes an artistic and intuitive way to represent AI by integrating visualization, sonification, and direct manipulation in VR, potentially targeting a wide range of audiences.

Bryan Wang, Mengyu Yang, Tovi Grossman

Learning musical instruments using online instructional videos has become increasingly prevalent. However, pre-recorded videos lack the instantaneous feedback and personal tailoring that human tutors provide. In addition, existing video navigations are not optimized for instrument learning, making the learning experience encumbered. Guided by our formative interviews with guitar players and prior literature, we designed Soloist, a mixed-initiative learning framework that automatically generates customizable curriculums from off-the-shelf guitar video lessons. Soloist takes raw videos as input and leverages deep-learning based audio processing to extract musical information. This back-end processing is used to provide an interactive visualization to support effective video navigation and real-time feedback on the user's performance, creating a guided learning experience. We demonstrate the capabilities and specific use-cases of Soloist within the domain of learning electric guitar solos using instructional YouTube videos. A remote user study, conducted to gather feedback from guitar players, shows encouraging results as the users unanimously preferred learning with Soloist over unconverted instructional videos.