An official website of the United States government
This content will become publicly available on January 19, 2026
A High-Quality Text-Rich Image Instruction Tuning Dataset via Hybrid Instruction Generation
- Award ID(s):
- 2223292
- PAR ID:
- 10637263
- Publisher / Repository:
- Proceedings of the 31st International Conference on Computational Linguistics
- Date Published:
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
In this article, we overview a professional learn- ing task that involves drawing one’s vision for high-quality, equitable mathematics instruction (HQEMI). The task is part of the ongoing work of a statewide research practice partnership that supports a shared vision of mathematics across the state K–12 system. Our work of HQEMI is rooted in the development of Munter’s (2014) four dimensions for visions of high-quality mathematics instruction (VHQMI): the role of the teacher, classroom discourse, mathematical tasks, and student engagement. The first three dimensions are particularly useful in the work of the drawing task. In this article, we share an overview of the drawing task, its implementa- tion with educators, and sample drawings, de- tailing how personal drawings were made visible across participants and the conversations result- ing from viewing and reflecting on one another’s drawings. These conversations helped surface disparities in notions of ideal mathematics in- struction and provided space for negotiation of shared meaning. We provide themes and over- arching considerations from these conversations to highlight discussions that might be elicited through this task in future iterations. Finally, we provide recommendations for implementing the task and consider how the task might be adapt- ed for others’ contexts to support professional learning about and development of a shared vision for mathematics.more » « less
-
null (Ed.)The Introduction to engineering (EGGN-100) is a project-based course offered every fall semester to first-year students with undecided engineering majors at California State University, Fullerton (CSUF). The primary objective of this course is to provide project-based learning (PBL) and introduce these students to major projects in Civil, Mechanical, Electrical, and Computer Engineering projects so that they can make an informed decision about their major. The PBL is an active learning method that aims to engage students in acquiring knowledge and skills through real-world experiences and well-planned project activities in engineering disciplines. The course comprises four team-based unique projects related to Civil, Mechanical, Electrical, and Computer Engineering. The project involves using a variety of engineering tools like AutoCAD, Multisim, and Arduino platforms. For the first time, due to the COVID-19 pandemic, the hands-on project-based EGGN-100 course was offered virtually. In this research, we document the learning experiences of students who attended EGGN-100 in a traditional face-to-face mode of instruction and students who participated in the same course in a virtual instruction mode. Surveys conducted during seemingly different modes of instruction show varying levels of satisfaction among students. Of the students who attended the course in traditional and instructional instruction mode, 69% and 90% responded that discipline-specific projects enabled them to make an informed decision, and PBL helped them choose their preferred major. Even the percentage of students who believed the PBL helped them make an informed decision about their major, they like to do more hands-on projects and prefer to attend the classes on campus. Students rated higher satisfaction in virtual instructional mode primarily due to the availability of video lectures, self-paced learning, and readily accessible project simulations. Learning by doing would have bought out the challenges and minor nuances of designing and executing an engineering project. Learning by watching is surficial and not necessarily exposes students to minor details that are critical. As such, the significance of this study is that maybe, after all, not all courses can be taught in a virtual environment, and some courses may be strictly taught in a traditional, hands-on instruction mode. We also study the socio-psychological impact of traditional and virtual learning experiences and report the remedies to cope with stress and loneliness in the online learning environment.more » « less
-
Modern server workloads have large code footprints which are prone to front-end bottlenecks due to instruction cache capacity misses. Even with the aggressive fetch directed instruction prefetching (FDIP), implemented in modern processors, there are still significant front-end stalls due to I-Cache misses. A major portion of misses that occur on a BPU-predicted path are tolerated by FDIP without causing stalls. Prior work on instruction prefetching, however, has not been designed to work with FDIP processors. Their singular goal is reducing I-Cache misses, whereas FDIP processors are designed to tolerate them. Designing an instruction prefetcher that works in conjunction with FDIP requires identifying the fraction of cache misses that impact front-end performance (that are not fully hidden by FDIP), and only targeting them. In this paper, we propose Priority Directed Instruction Prefetching (PDIP), a novel instruction prefetching technique that complements FDIP by issuing prefetches for only targets where FDIP struggles - along the resteer path of front-end stall-causing events. PDIP identifies these targets and associates them with a trigger for future prefetch. At a 43.5KB budget, PDIP achieves up to 5.1% IPC speedup on important workloads such as cassandra and a geomean IPC speedup of 3.2% across 16 benchmarks.more » « less
