Author name cluster

Joy Hsu

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

12 papers

2 author rows

AAAI Conference 2026 Conference Paper

Discovering Hybrid World Representations with Co-Evolving Foundation Models

Jiajun Wu
Yunzhi Zhang
Hong-Xing Yu
Joy Hsu
Jiayuan Mao

This perspective article discusses an emerging research direction: to what extent can foundation models yield usable structure for modeling the physical world? We offer a Markovian formulation of structured world models and outline the notion of multi-level hybrid world representations that support compositional structure. We then review and suggest possible discovery paradigms, spanning distillation, interaction-driven continual learning, and ensemble learning.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

From Programs to Poses: Factored Real-World Scene Generation via Learned Program Libraries

Joy Hsu
Emily Jin
Jiajun Wu
Niloy Mitra

Real-world scenes, such as those in ScanNet, are difficult to capture, with highly limited data available. Generating realistic scenes with varied object poses remains an open and challenging task. In this work, we propose FactoredScenes, a framework that synthesizes realistic 3D scenes by leveraging the underlying structure of rooms while learning the variation of object poses from lived-in scenes. We introduce a factored representation that decomposes scenes into hierarchically organized concepts of room programs and object poses. To encode structure, FactoredScenes learns a library of functions capturing reusable layout patterns from which scenes are drawn, then uses large language models to generate high-level programs, regularized by the learned library. To represent scene variations, FactoredScenes learns a program-conditioned model to hierarchically predict object poses, and retrieves and places 3D objects in a scene. We show that FactoredScenes generates realistic, real-world rooms that are difficult to distinguish from real ScanNet scenes.

PDF Details

ICLR Conference 2025 Conference Paper

Predicate Hierarchies Improve Few-Shot State Classification

Emily Jin
Joy Hsu
Jiajun Wu 0001

State classification of objects and their relations is core to many long-horizon tasks, particularly in robot planning and manipulation. However, the combinatorial explosion of possible object-predicate combinations, coupled with the need to adapt to novel real-world environments, makes it a desideratum for state classification models to generalize to novel queries with few examples. To this end, we propose PHIER, which leverages predicate hierarchies to generalize effectively in few-shot scenarios. PHIER uses an object-centric scene encoder, self-supervised losses that infer semantic relations between predicates, and a hyperbolic distance metric that captures hierarchical structure; it learns a structured latent space of image-predicate pairs that guides reasoning over state classification queries. We evaluate PHIER in the CALVIN and BEHAVIOR robotic environments and show that PHIER significantly outperforms existing methods in few-shot, out-of-distribution state classification, and demonstrates strong zero- and few-shot generalization from simulated to real-world tasks. Our results demonstrate that leveraging predicate hierarchies improves performance on state classification tasks with limited data.

Details

TMLR Journal 2025 Journal Article

Visually Descriptive Language Model for Vector Graphics Reasoning

Zhenhailong Wang
Joy Hsu
Xingyao Wang
Kuan-Hao Huang
Manling Li
Jiajun Wu
Heng Ji

Despite significant advancements, current large multimodal models (LMMs) struggle to bridge the gap between low-level visual perception—focusing on shapes, sizes, and layouts—and high-level language reasoning involving semantics, events, and logic. This limitation becomes evident in tasks requiring precise visual perception, such as comparing geometric properties or solving visual algorithmic reasoning problems. To study this failure mode, we focus on an important visual domain: vector graphics —images composed purely of 2D objects and shapes, which are prevalent in Web, PC, and Mobile environments. Importantly, we consider rasterized vector graphics without assuming access to their underlying vector code. We identify two key research questions: how can we enable precise visual perception, and how can we facilitate high-level reasoning based on such low-level perceptions? To accurately capture low-level visual details, we explore using SVG for the precise encoding of visual scenes. However, SVGs are not readily interpretable by LLMs or LMMs in a zero-shot manner. To address this challenge, we propose the Visually Descriptive Language Model (VDLM) to build a bridge between low-level visual perception and high-level language reasoning. VDLM learns an intermediate symbolic representation called Primal Visual Description (PVD), which translates raw SVGs into a higher-level abstraction comprising primitive attributes. This abstraction allows for direct interpretation by foundation models for zero-shot generalization to different reasoning tasks. Without any human-annotated data, VDLM leads to significant improvements in state-of-the-art LMMs, such as GPT-4o, across various low-level multimodal perception and reasoning tasks on rasterized vector graphics. Additionally, we provide extensive analyses of VDLM’s performance, showing that our framework offers improved interpretability due to its disentangled perception and reasoning processes. As the first attempt to construct a descriptive intermediate representation for low-level visual reasoning, we also conduct an in-depth error analysis, highlighting remaining limitations and suggesting directions for future research.

PDF Details

ICLR Conference 2025 Conference Paper

What Makes a Maze Look Like a Maze?

Joy Hsu
Jiayuan Mao
Joshua B. Tenenbaum
Noah D. Goodman
Jiajun Wu 0001

A unique aspect of human visual understanding is the ability to flexibly interpret abstract concepts: acquiring lifted rules explaining what they symbolize, grounding them across familiar and unfamiliar contexts, and making predictions or reasoning about them. While off-the-shelf vision-language models excel at making literal interpretations of images (e.g., recognizing object categories such as tree branches), they still struggle to make sense of such visual abstractions (e.g., how an arrangement of tree branches may form the walls of a maze). To address this challenge, we introduce Deep Schema Grounding (DSG), a framework that leverages explicit structured representations of visual abstractions for grounding and reasoning. At the core of DSG are schemas—dependency graph descriptions of abstract concepts that decompose them into more primitive-level symbols. DSG uses large language models to extract schemas, then hierarchically grounds concrete to abstract components of the schema onto images with vision-language models. The grounded schema is used to augment visual abstraction understanding. We systematically evaluate DSG and different methods in reasoning on our new Visual Abstractions Benchmark, which consists of diverse, real-world images of abstract concepts and corresponding question-answer pairs labeled by humans. We show that DSG significantly improves the abstract visual reasoning performance of vision-language models, and is a step toward human-aligned understanding of visual abstractions.

Details

ICLR Conference 2024 Conference Paper

Learning Planning Abstractions from Language

Weiyu Liu
Geng Chen
Joy Hsu
Jiayuan Mao
Jiajun Wu 0001

This paper presents a framework for learning state and action abstractions in sequential decision-making domains. Our framework, planning abstraction from language (PARL), utilizes language-annotated demonstrations to automatically discover a symbolic and abstract action space and induce a latent state abstraction based on it. PARL consists of three stages: 1) recovering object-level and action concepts, 2) learning state abstractions, abstract action feasibility, and transition models, and 3) applying low-level policies for abstract actions. During inference, given the task description, PARL first makes abstract action plans using the latent transition and feasibility functions, then refines the high-level plan using low-level policies. PARL generalizes across scenarios involving novel object instances and environments, unseen concept compositions, and tasks that require longer planning horizons than settings it is trained on.

Details

TMLR Journal 2023 Journal Article

DisCo: Improving Compositional Generalization in Visual Reasoning through Distribution Coverage

Joy Hsu
Jiayuan Mao
Jiajun Wu

We present DisCo, a learning paradigm for improving compositional generalization of visual reasoning models by leveraging unlabeled, out-of-distribution images from the test distribution. DisCo has two components. The first is an iterative pseudo-labeling framework with an entropy measure, which effectively labels images of novel attribute compositions paired with randomly sampled questions. The second is a distribution coverage metric, serving as a model selection strategy that approximates generalization capability to test examples drawn from a different attribute combination distribution to the train set, without the use of labeled data from the test distribution. Both components are built on strong empirical evidence of the correlation between the chosen metric and model generalization, and improve distribution coverage on unlabeled images. We apply DisCo to visual question answering, with three backbone networks (FiLM, TbD-net, and the Neuro-Symbolic Concept Learner), and demonstrate that it consistently enhances performance on a variety of compositional generalization tasks with varying levels of train data bias.

PDF Details

ICML Conference 2023 Conference Paper

Motion Question Answering via Modular Motion Programs

Mark Endo
Joy Hsu
Jiaman Li
Jiajun Wu 0001

In order to build artificial intelligence systems that can perceive and reason with human behavior in the real world, we must first design models that conduct complex spatio-temporal reasoning over motion sequences. Moving towards this goal, we propose the HumanMotionQA task to evaluate complex, multi-step reasoning abilities of models on long-form human motion sequences. We generate a dataset of question-answer pairs that require detecting motor cues in small portions of motion sequences, reasoning temporally about when events occur, and querying specific motion attributes. In addition, we propose NSPose, a neuro-symbolic method for this task that uses symbolic reasoning and a modular design to ground motion through learning motion concepts, attribute neural operators, and temporal relations. We demonstrate the suitability of NSPose for the HumanMotionQA task, outperforming all baseline methods.

Details

ICLR Conference 2023 Conference Paper

Programmatically Grounded, Compositionally Generalizable Robotic Manipulation

Renhao Wang
Jiayuan Mao
Joy Hsu
Hang Zhao 0021
Jiajun Wu 0001
Yang Gao 0029

Robots operating in the real world require both rich manipulation skills as well as the ability to semantically reason about when to apply those skills. Towards this goal, recent works have integrated semantic representations from large-scale pretrained vision-language (VL) models into manipulation models, imparting them with more general reasoning capabilities. However, we show that the conventional {\it pretraining-finetuning} pipeline for integrating such representations entangles the learning of domain-specific action information and domain-general visual information, leading to less data-efficient training and poor generalization to unseen objects and tasks. To this end, we propose \ours, a {\it modular} approach to better leverage pretrained VL models by exploiting the syntactic and semantic structures of language instructions. Our framework uses a semantic parser to recover an executable program, composed of functional modules grounded on vision and action across different modalities. Each functional module is realized as a combination of deterministic computation and learnable neural networks. Program execution produces parameters to general manipulation primitives for a robotic end-effector. The entire modular network can be trained with end-to-end imitation learning objectives. Experiments show that our model successfully disentangles action and perception, translating to improved zero-shot and compositional generalization in a variety of manipulation behaviors. Project webpage at: \url{https://progport.github.io}.

Details

NeurIPS Conference 2023 Conference Paper

What’s Left? Concept Grounding with Logic-Enhanced Foundation Models

Joy Hsu
Jiayuan Mao
Josh Tenenbaum
Jiajun Wu

Recent works such as VisProg and ViperGPT have smartly composed foundation models for visual reasoning—using large language models (LLMs) to produce programs that can be executed by pre-trained vision-language models. However, they operate in limited domains, such as 2D images, not fully exploiting the generalization of language: abstract concepts like “ left ” can also be grounded in 3D, temporal, and action data, as in moving to your left. This limited generalization stems from these inference-only methods’ inability to learn or adapt pre-trained models to a new domain. We propose the L ogic- E nhanced F ounda T ion Model ( LEFT ), a unified framework that learns to ground and reason with concepts across domains with a differentiable, domain-independent, first-order logic-based program executor. LEFT has an LLM interpreter that outputs a program represented in a general, logic-based reasoning language, which is shared across all domains and tasks. LEFT’s executor then executes the program with trainable domain-specific grounding modules. We show that LEFT flexibly learns concepts in four domains: 2D images, 3D scenes, human motions, and robotic manipulation. It exhibits strong reasoning ability in a wide variety of tasks, including those that are complex and not seen during training, and can be easily applied to new domains.

PDF Details

NeurIPS Conference 2022 Conference Paper

Geoclidean: Few-Shot Generalization in Euclidean Geometry

Joy Hsu
Jiajun Wu
Noah Goodman

Euclidean geometry is among the earliest forms of mathematical thinking. While the geometric primitives underlying its constructions, such as perfect lines and circles, do not often occur in the natural world, humans rarely struggle to perceive and reason with them. Will computer vision models trained on natural images show the same sensitivity to Euclidean geometry? Here we explore these questions by studying few-shot generalization in the universe of Euclidean geometry constructions. We introduce Geoclidean, a domain-specific language for Euclidean geometry, and use it to generate two datasets of geometric concept learning tasks for benchmarking generalization judgements of humans and machines. We find that humans are indeed sensitive to Euclidean geometry and generalize strongly from a few visual examples of a geometric concept. In contrast, low-level and high-level visual features from standard computer vision models pretrained on natural images do not support correct generalization. Thus Geoclidean represents a novel few-shot generalization benchmark for geometric concept learning, where the performance of humans and of AI models diverge. The Geoclidean framework and dataset are publicly available for download.

PDF Details

NeurIPS Conference 2021 Conference Paper

Capturing implicit hierarchical structure in 3D biomedical images with self-supervised hyperbolic representations

Joy Hsu
Jeffrey Gu
Gong Wu
Wah Chiu
Serena Yeung

We consider the task of representation learning for unsupervised segmentation of 3D voxel-grid biomedical images. We show that models that capture implicit hierarchical relationships between subvolumes are better suited for this task. To that end, we consider encoder-decoder architectures with a hyperbolic latent space, to explicitly capture hierarchical relationships present in subvolumes of the data. We propose utilizing a 3D hyperbolic variational autoencoder with a novel gyroplane convolutional layer to map from the embedding space back to 3D images. To capture these relationships, we introduce an essential self-supervised loss---in addition to the standard VAE loss---which infers approximate hierarchies and encourages implicitly related subvolumes to be mapped closer in the embedding space. We present experiments on synthetic datasets along with a dataset from the medical domain to validate our hypothesis.

PDF Details