Albert Lu

Albert Lu, Hongxin Zhang, Yanzhe Zhang et al. · gatech

The limits of open-ended generative models are unclear, yet increasingly important. What causes them to succeed and what causes them to fail? In this paper, we take a prompt-centric approach to analyzing and bounding the abilities of open-ended generative models. We present a generic methodology of analysis with two challenging prompt constraint types: structural and stylistic. These constraint types are categorized into a set of well-defined constraints that are analyzable by a single prompt. We then systematically create a diverse set of simple, natural, and useful prompts to robustly analyze each individual constraint. Using the GPT-3 text-davinci-002 model as a case study, we generate outputs from our collection of prompts and analyze the model's generative failures. We also show the generalizability of our proposed method on other large models like BLOOM and OPT. Our results and our in-context mitigation strategies reveal open challenges for future research. We have publicly released our code at https://github.com/SALT-NLP/Bound-Cap-LLM.

Albert Lu, Jordan Marshall, Yifan Wang et al.

In this paper, two methodologies are used to speed up the maximization of the breakdown volt-age (BV) of a vertical GaN diode that has a theoretical maximum BV of ~2100V. Firstly, we demonstrated a 5X faster accurate simulation method in Technology Computer-Aided-Design (TCAD). This allows us to find 50% more numbers of high BV (>1400V) designs at a given simulation time. Secondly, a machine learning (ML) model is developed using TCAD-generated data and used as a surrogate model for differential evolution optimization. It can inversely design an out-of-the-training-range structure with BV as high as 1887V (89% of the ideal case) compared to ~1100V designed with human domain expertise.

Thomas Lu, Albert Lu, Hiu Yung Wong

This paper demonstrates the learning of the underlying device physics by mapping device structure images to their corresponding Current-Voltage (IV) characteristics using a novel framework based on variational autoencoders (VAE). Since VAE is used, domain expertise is not required and the framework can be quickly deployed on any new device and measurement. This is expected to be useful in the compact modeling of novel devices when only device cross-sectional images and electrical characteristics are available (e.g. novel emerging memory). Technology Computer-Aided Design (TCAD) generated and hand-drawn Metal-Oxide-Semiconductor (MOS) device images and noisy drain-current-gate-voltage curves (IDVG) are used for the demonstration. The framework is formed by stacking two VAEs (one for image manifold learning and one for IDVG manifold learning) which communicate with each other through the latent variables. Five independent variables with different strengths are used. It is shown that it can perform inverse design (generate a design structure for a given IDVG) and forward prediction (predict IDVG for a given structure image, which can be used for compact modeling if the image is treated as device parameters) successfully. Since manifold learning is used, the machine is shown to be robust against noise in the inputs (i.e. using hand-drawn images and noisy IDVG curves) and not confused by weak and irrelevant independent variables.

Albert Lu, Meng Jiang

Review score prediction requires review text understanding, a critical real-world application of natural language processing. Due to dissimilar text domains in product reviews, a common practice is fine-tuning BERT models upon reviews of differing domains. However, there has not yet been an empirical study of cross-domain behaviors of BERT models in the various tasks of product review understanding. In this project, we investigate text classification BERT models fine-tuned on single-domain and multi-domain Amazon review data. In our findings, though single-domain models achieved marginally improved performance on their corresponding domain compared to multi-domain models, multi-domain models outperformed single-domain models when evaluated on multi-domain data, single-domain data the single-domain model was not fine-tuned on, and on average when considering all tests. Though slight increases in accuracy can be achieved through single-domain model fine-tuning, computational resources and costs can be reduced by utilizing multi-domain models that perform well across domains.

Albert Lu, Yu Foon Chau, Hiu Yung Wong

Machine learning (ML) is promising in assisting technology computer-aided design (TCAD) simulations to alleviate difficulty in convergence and prolonged simulation time. While ML is widely used in TCAD, they either require access to the internal solver, require extensive domain expertise, are only trained by terminal quantities such as currents and voltages, and/or lack out-of-training-range prediction capability. In this paper, using Si nanowire as an example, we demonstrate that it is possible to use a physics-informed neural network (PINN) to predict out-of-training-range TCAD solutions without accessing the internal solver and with minimal domain expertise. The machine not only can predict a 2.5 times larger range than the training but also can predict the inversion region by only being trained with subthreshold region data. The physics-informed module is also trained with data without the need for human-coded equations making this easier to be extended to more sophisticated systems.

Iddo Drori, Sarah Zhang, Reece Shuttleworth et al.

We demonstrate that a neural network pre-trained on text and fine-tuned on code solves mathematics course problems, explains solutions, and generates new questions at a human level. We automatically synthesize programs using few-shot learning and OpenAI's Codex transformer and execute them to solve course problems at 81% automatic accuracy. We curate a new dataset of questions from MIT's largest mathematics courses (Single Variable and Multivariable Calculus, Differential Equations, Introduction to Probability and Statistics, Linear Algebra, and Mathematics for Computer Science) and Columbia University's Computational Linear Algebra. We solve questions from a MATH dataset (on Prealgebra, Algebra, Counting and Probability, Intermediate Algebra, Number Theory, and Precalculus), the latest benchmark of advanced mathematics problems designed to assess mathematical reasoning. We randomly sample questions and generate solutions with multiple modalities, including numbers, equations, and plots. The latest GPT-3 language model pre-trained on text automatically solves only 18.8% of these university questions using zero-shot learning and 30.8% using few-shot learning and the most recent chain of thought prompting. In contrast, program synthesis with few-shot learning using Codex fine-tuned on code generates programs that automatically solve 81% of these questions. Our approach improves the previous state-of-the-art automatic solution accuracy on the benchmark topics from 8.8% to 81.1%. We perform a survey to evaluate the quality and difficulty of generated questions. This work is the first to automatically solve university-level mathematics course questions at a human level and the first work to explain and generate university-level mathematics course questions at scale, a milestone for higher education.