We are teaching a major update of the course Spring 2021 as an official UC Berkeley course and as an online course, with all lectures and labs available for free. There are many great courses to learn how to train deep neural networks. However, training the model is just one part of shipping a deep learning project. The course is aimed at people who already know the basics of deep learning and want to understand the rest of the process of creating production deep learning systems. While we cover the basics of deep learning (backpropagation, convolutional neural networks, recurrent neural networks, transformers, etc), we expect these lectures to be mostly review.

The week of February 15, we cover RNNs and applications in Natural Language Processing, and start doing sequence processing in lab.

In this section we will learn - What does Machine Learning mean. What are the meanings or different terms associated with machine learning? You will see some examples so that you understand what machine learning actually is. It also contains steps involved in building a machine learning model, not just linear models, any machine learning model.

Open Source Software projects add labels to open issues to help contributors choose tasks. However, manually labeling issues is time-consuming and error-prone. Current automatic approaches for creating labels are mostly limited to classifying issues as a bug/non-bug. In this paper, we investigate the feasibility and relevance of labeling issues with the domain of the APIs required to complete the tasks. We leverage the issues' description and the project history to build prediction models, which resulted in precision up to 82% and recall up to 97.8%. We also ran a user study (n=74) to assess these labels' relevancy to potential contributors. The results show that the labels were useful to participants in choosing tasks, and the API-domain labels were selected more often than the existing architecture-based labels. Our results can inspire the creation of tools to automatically label issues, helping developers to find tasks that better match their skills.

We study an interesting and challenging problem, learning any part of a Bayesian network (BN) structure. In this challenge, it will be computationally inefficient using existing global BN structure learning algorithms to find an entire BN structure to achieve the part of a BN structure in which we are interested. And local BN structure learning algorithms encounter the false edge orientation problem when they are directly used to tackle this challenging problem. In this paper, we first present a new concept of Expand-Backtracking to explain why local BN structure learning methods have the false edge orientation problem, then propose APSL, an efficient and accurate Any Part of BN Structure Learning algorithm. Specifically, APSL divides the V-structures in a Markov blanket (MB) into two types: collider V-structure and non-collider V-structure, then it starts from a node of interest and recursively finds both collider V-structures and non-collider V-structures in the found MBs, until the part of a BN structure in which we are interested are oriented. To improve the efficiency of APSL, we further design the APSL-FS algorithm using Feature Selection, APSL-FS. Using six benchmark BNs, the extensive experiments have validated the efficiency and accuracy of our methods.

Empirical Risk Minimization (ERM) based machine learning algorithms have suffered from weak generalization performance on data obtained from out-of-distribution (OOD). To address this problem, Invariant Risk Minimization (IRM) objective was suggested to find invariant optimal predictor which is less affected by the changes in data distribution. However, even with such progress, IRMv1, the practical formulation of IRM, still shows performance degradation when there are not enough training data, and even fails to generalize to OOD, if the number of spurious correlations is larger than the number of environments. In this paper, to address such problems, we propose a novel meta-learning based approach for IRM. In this method, we do not assume the linearity of classifier for the ease of optimization, and solve ideal bi-level IRM objective with Model-Agnostic Meta-Learning (MAML) framework. Our method is more robust to the data with spurious correlations and can provide an invariant optimal classifier even when data from each distribution are scarce. In experiments, we demonstrate that our algorithm not only has better OOD generalization performance than IRMv1 and all IRM variants, but also addresses the weakness of IRMv1 with improved stability.

If an unknown example that is not seen during training appears, most recognition systems usually produce overgeneralized results and determine that the example belongs to one of the known classes. To address this problem, teacher-explorer-student (T/E/S) learning, which adopts the concept of open set recognition (OSR) that aims to reject unknown samples while minimizing the loss of classification performance on known samples, is proposed in this study. In this novel learning method, overgeneralization of deep learning classifiers is significantly reduced by exploring various possibilities of unknowns. Here, the teacher network extracts some hints about unknowns by distilling the pretrained knowledge about knowns and delivers this distilled knowledge to the student. After learning the distilled knowledge, the student network shares the learned information with the explorer network. Then, the explorer network shares its exploration results by generating unknown-like samples and feeding the samples to the student network. By repeating this alternating learning process, the student network experiences a variety of synthetic unknowns, reducing overgeneralization. Extensive experiments were conducted, and the experimental results showed that each component proposed in this paper significantly contributes to the improvement in OSR performance. As a result, the proposed T/E/S learning method outperformed current state-of-the-art methods.

The rise of language models such as BERT allows for high-quality text paraphrasing. This is a problem to academic integrity, as it is difficult to differentiate between original and machine-generated content. We propose a benchmark consisting of paraphrased articles using recent language models relying on the Transformer architecture. Our contribution fosters future research of paraphrase detection systems as it offers a large collection of aligned original and paraphrased documents, a study regarding its structure, classification experiments with state-of-the-art systems, and we make our findings publicly available.

Case Management has been gradually evolving to support Knowledge-intensive business process management, which resulted in developing different modeling languages, e.g., Declare, Dynamic Condition Response (DCR), and Case Management Model and Notation (CMMN). A language will die if users do not accept and use it in practice - similar to extinct human languages. Thus, it is important to evaluate how users perceive languages to determine if there is a need for improvement. Although some studies have investigated how the process designers perceived Declare and DCR, there is a lack of research on how they perceive CMMN. Therefore, this study investigates how the process designers perceive the usefulness and ease of use of CMMN and DCR based on the Technology Acceptance Model. DCR is included to enable comparing the study result with previous ones. The study is performed by educating master level students with these languages over eight weeks by giving feedback on their assignments to reduce perceptions biases. The students' perceptions are collected through questionnaires before and after sending feedback on their final practice in the exam. Thus, the result shows how the perception of participants can change by receiving feedback - despite being well trained. The reliability of responses is tested using Cronbach's alpha, and the result indicates that both languages have an acceptable level for both perceived usefulness and ease of use.

Today, AI and ML (Artificial Intelligence, Machine Learning) are creating havoc in a number of fields and industries, including education. As per report of Ideamotive, according to Market Research Engine, the global AI in the education market will reach $5.80 billion by 2025 at a compound annual growth rate of 45%. The use of machine learning and artificial intelligence has assisted in the development of a more effective and time-saving approach for teachers to follow a clear ride of guidance. This has aided students' comprehension and has gone beyond the criteria of human intelligence. Many ed-tech companies have started to deploy digital programs to boost the educational experiences, as AI in ed-tech helps respond to students' specific needs. It helps students in assessing their success on various subjects and keeps track of their input.