Max Zuo

Ali Mirzazadeh, Florian Dubost, Maxwell Pike et al. · stanford

Attention--or attribution--maps methods are methods designed to highlight regions of the model's input that were discriminative for its predictions. However, different attention maps methods can highlight different regions of the input, with sometimes contradictory explanations for a prediction. This effect is exacerbated when the training set is small. This indicates that either the model learned incorrect representations or that the attention maps methods did not accurately estimate the model's representations. We propose an unsupervised fine-tuning method that optimizes the consistency of attention maps and show that it improves both classification performance and the quality of attention maps. We propose an implementation for two state-of-the-art attention computation methods, Grad-CAM and Guided Backpropagation, which relies on an input masking technique. We also show results on Grad-CAM and Integrated Gradients in an ablation study. We evaluate this method on our own dataset of event detection in continuous video recordings of hospital patients aggregated and curated for this work. As a sanity check, we also evaluate the proposed method on PASCAL VOC and SVHN. With the proposed method, with small training sets, we achieve a 6.6 points lift of F1 score over the baselines on our video dataset, a 2.9 point lift of F1 score on PASCAL, and a 1.8 points lift of mean Intersection over Union over Grad-CAM for weakly supervised detection on PASCAL. Those improved attention maps may help clinicians better understand vision model predictions and ease the deployment of machine learning systems into clinical care. We share part of the code for this article at the following repository: https://github.com/alimirzazadeh/SemisupervisedAttention.

Reza Esfandiarpoor, Max Zuo, Stephen H. Bach

We introduce Trove, an easy-to-use open-source retrieval toolkit that simplifies research experiments without sacrificing flexibility or speed. For the first time, we introduce efficient data management features that load and process (filter, select, transform, and combine) retrieval datasets on the fly, with just a few lines of code. This gives users the flexibility to easily experiment with different dataset configurations without the need to compute and store multiple copies of large datasets. Trove is highly customizable: in addition to many built-in options, it allows users to freely modify existing components or replace them entirely with user-defined objects. It also provides a low-code and unified pipeline for evaluation and hard negative mining, which supports multi-node execution without any code changes. Trove's data management features reduce memory consumption by a factor of 2.6. Moreover, Trove's easy-to-use inference pipeline incurs no overhead, and inference times decrease linearly with the number of available nodes. Most importantly, we demonstrate how Trove simplifies retrieval experiments and allows for arbitrary customizations, thus facilitating exploratory research.

Max Zuo, Logan Schick, Matthew Gombolay et al.

Modern day computer games have extremely large state and action spaces. To detect bugs in these games' models, human testers play the games repeatedly to explore the game and find errors in the games. Such gameplay is exhaustive and time consuming. Moreover, since robotics simulators depend on similar methods of model specification and debugging, the problem of finding errors in the model is of interest to the robotics community to ensure robot behaviors and interactions are consistent in simulators. Previous methods have used reinforcement learning arXiv:2103.13798 and search based methods (Chang, 2019, (Chang, 2021) arXiv:1811.06962 including Rapidly-exploring Random Trees (RRT) to explore a game's state-action space to find bugs. However, such search and exploration based methods are not efficient at exploring the state-action space without a pre-defined heuristic. In this work we attempt to combine a human-tester's expertise in solving games, and the RRT's exhaustiveness to search a game's state space efficiently with high coverage. This paper introduces Cloning Assisted RRT (CA-RRT) to test a game through search. We compare our methods to two existing baselines: 1) a weighted-RRT as described by arXiv:1812.03125; 2) human demonstration seeded RRT as described by Chang et. al. We find CA-RRT is applicable to more game maps and explores more game states in fewer tree expansions/iterations when compared to the existing baselines. In each test, CA-RRT reached more states on average in the same number of iterations as weighted-RRT. In our tested environments, CA-RRT reached the same number of states as weighted-RRT by more than 5000 fewer iterations on average, almost a 50% reduction and applied to more scenarios than. Moreover, as a consequence of our first person behavior cloning approach, CA-RRT worked on unseen game maps than just seeding the RRT with human demonstrated states.

Max Zuo, Francisco Piedrahita Velez, Xiaochen Li et al.

Recent works have explored using language models for planning problems. One approach examines translating natural language descriptions of planning tasks into structured planning languages, such as the planning domain definition language (PDDL). Existing evaluation methods struggle to ensure semantic correctness and rely on simple or unrealistic datasets. To bridge this gap, we introduce \textit{Planetarium}, a benchmark designed to evaluate language models' ability to generate PDDL code from natural language descriptions of planning tasks. \textit{Planetarium} features a novel PDDL equivalence algorithm that flexibly evaluates the correctness of generated PDDL, along with a dataset of 145,918 text-to-PDDL pairs across 73 unique state combinations with varying levels of difficulty. Finally, we evaluate several API-access and open-weight language models that reveal this task's complexity. For example, 96.1\% of the PDDL problem descriptions generated by GPT-4o are syntactically parseable, 94.4\% are solvable, but only 24.8\% are semantically correct, highlighting the need for a more rigorous benchmark for this problem.

Yeganeh Kordi, Nihal V. Nayak, Max Zuo et al.

We investigate how well large language models (LLMs) generalize across different task difficulties, a key question for effective data curation and evaluation. Existing research is mixed regarding whether training on easier or harder data leads to better results, and whether those gains come on easier or harder test data. We address this question by conducting a systematic evaluation of LLMs' generalization across models, datasets, and fine-grained groups of example difficulty. We rank examples in six datasets using the outputs of thousands of different LLMs and Item Response Theory (IRT), a well-established difficulty metric in educational testing. Unlike prior work, our difficulty ratings are therefore determined solely by the abilities of many different LLMs, excluding human opinions of difficulty. With a more objective, larger-scale, and finer-grained analysis, we show that cross-difficulty generalization is often limited; training on either easy or hard data cannot achieve consistent improvements across the full range of difficulties. These results show the importance of having a range of difficulties in both training and evaluation data for LLMs, and that taking shortcuts with respect to difficulty is risky.