NEC Labs America is proud to be a Silver Sponsor for NeurIPS 2023 in New Orleans from December 10-16. Visit our booth 1122 to meet our team and learn about our intern opportunities in machine learning, data science, media analytics and integrated systems.
Exploring Question Decomposition for Zero-Shot VQA Paper
Our Vijay Kumar.B.G, and Zaid Khan, Northeastern University will present a paper, Exploring Question Decomposition for Zero-Shot VQA at NeurIPS 2023.
Abstract: Visual question answering (VQA) has traditionally been treated as a single-step task where each question receives the same amount of effort, unlike natural human question-answering strategies. We explore a question decomposition strategy for VQA to overcome this limitation. We probe the ability of recently developed large vision-language models to use human-written decompositions and produce their own decompositions of visual questions, finding they are capable of learning both tasks from demonstrations alone. However, we show that naive application of model-written decompositions can hurt performance. We introduce a model-driven selective decomposition approach for second-guessing predictions and correcting errors, and validate its effectiveness on eight VQA tasks across three domains, showing consistent improvements in accuracy, including improvements of >20% on medical VQA datasets and boosting the zero-shot performance of BLIP-2 above chance on a VQA reformulation of the challenging Winoground task.
Presented Papers Schedule
Event | Event ID | Presenter | Presenter Institution | Type | Date |
DP-Mix: Mixup-based Data Augmentation for Differentially Private Learning | 72682 | Francesco Pittaluga |
NEC Laboratories America | Poster | 2023-12-12 15:15:00 |
Curriculum Learning for Graph Neural Networks: Which Edges Should We Learn First | 70907 | Junxiang Wang |
NEC Laboratories America | Poster | 2023-12-14 8:45:00 |
Exploring Question Decomposition for Zero-Shot VQA | 71957 | Vijay Kumar B G and Zaid Khan | NEC Laboratories America | Poster | 2023-12-13 8:45:00 |
Hierarchical Gaussian Mixture based Task Generative Model for Robust Meta-Learning | 70550 | Wei Cheng |
NEC Laboratories America | Poster | 2023-12-14 8:45:00 |
Hierarchical Gaussian Mixture based Task Generative Model for Robust Meta-Learning | 70550 | Zhengzhang Chen |
NEC Laboratories America | Poster | 2023-12-14 8:45:00 |
Disentangled Wasserstein Autoencoder for T-Cell Receptor Engineering | 72313 | Presenter Tianxiao Li | NEC Laboratories America | Poster | 2023-12-12 8:45:00 |
Hierarchical Gaussian Mixture based Task Generative Model for Robust Meta-Learning | 70550 | Haifeng Chen |
NEC Laboratories America | Poster | 2023-12-14 8:45:00 |
Achieving Counterfactual Fairness in Changing Environments via Sequential Autoencoder | 78067 | Haifeng Chen |
NEC Laboratories America | ||
Weakly-Supervised Concealed Object Segmentation with SAM-based Pseudo Labeling and Multi-scale Feature Grouping | Presenter Kai Li†, Chunming He†, Yachao Zhang, Guoxia Xu, Longxiang Tang, Yulun Zhang, Zhenhua Guo, Xiu Li. | NEC Laboratories America | 12/13/2023 11:45 | ||
NeurIPS 2023 Workshop Mathematics of Modern Machine Learning (M3L)
A Theoretical Study of Dataset Distillation |
Presenter Zach Izzo | NEC Laboratories America | 12/16/2023 |
Our Data Science & System Security Research
We aim to build novel big-data solutions and service platforms to simplify complex systems management. We develop new information technology that supports innovative applications, from big data analytics to the Internet of Things. Our experimental and theoretical research includes many data science and systems research domains. These include but are not limited to time series mining, deep learning, NLP and large language models, graph mining, signal processing, and cloud computing.