We study the design decisions of publicly available instruction tuning methods, and break down the development of Flan 2022 models (Chung et al., 2022). Through careful ablation studies on the Flan Collection of instruction tuning tasks and methods, we tease apart the effect of design decisions that enable Flan-T5 to outperform prior work by 3-17%+ across evaluation settings. We find task balancing and enrichment techniques are overlooked but critical to effective instruction tuning, and in particular, training with mixed prompt settings (zero-shot, few-shot, and chain-of-thought) actually yields stronger (2%+) performance in all settings. In further experiments, we show Flan-T5 requires less finetuning to converge higher and faster than T5 on single downstream tasks--motivating instruction-tuned models as more computationallyefficient starting checkpoints for new tasks. Finally, to accelerate research on instruction tuning, we make the Flan 2022 collection of datasets, templates, and methods publicly available.

Large Language Models (LLMs), such as ChatGPT, are quickly advancing AI to the frontiers of practical consumer use and leading industries to re-evaluate how they allocate resources for content production. Authoring of open educational resources and hint content within adaptive tutoring systems is labor intensive. Should LLMs like ChatGPT produce educational content on par with human-authored content, the implications would be significant for further scaling of computer tutoring system approaches. In this paper, we conduct the first learning gain evaluation of ChatGPT by comparing the efficacy of its hints with hints authored by human tutors with 77 participants across two algebra topic areas, Elementary Algebra and Intermediate Algebra. We find that 70% of hints produced by ChatGPT passed our manual quality checks and that both human and ChatGPT conditions produced positive learning gains. However, gains were only statistically significant for human tutor created hints. Learning gains from human-created hints were substantially and statistically significantly higher than ChatGPT hints in both topic areas, though ChatGPT participants in the Intermediate Algebra experiment were near ceiling and not even with the control at pre-test. We discuss the limitations of our study and suggest several future directions for the field. Problem and hint content used in the experiment is provided for replicability.

Intelligent systems deployed in the real world suffer from catastrophic forgetting when exposed to a sequence of tasks. Humans, on the other hand, acquire, consolidate, and transfer knowledge between tasks that rarely interfere with the consolidated knowledge. Accompanied by self-regulated neurogenesis, continual learning in the brain is governed by a rich set of neurophysiological processes that harbor different types of knowledge, which are then integrated by conscious processing. Thus, inspired by the Global Workspace Theory of conscious information access in the brain, we propose TAMiL, a continual learning method that entails task-attention modules to capture task-specific information from the common representation space. We employ simple, undercomplete autoencoders to create a communication bottleneck between the common representation space and the global workspace, allowing only the task-relevant information to the global workspace, thus greatly reducing task interference. Experimental results show that our method outperforms state-of-the-art rehearsal-based and dynamic sparse approaches and bridges the gap between fixed capacity and parameter isolation approaches while being scalable. Deep neural networks (DNNs) deployed in the real world are normally required to learn multiple tasks sequentially and are exposed to non-stationary data distributions. Throughout their lifespan, such systems must acquire new skills without compromising previously learned knowledge. However, continual learning (CL) over multiple tasks violates the i.i.d. The menace of catastrophic forgetting occurs due to the stability-plasticity dilemma: the extent to which the system must be stable to retain consolidated knowledge and be plastic to assimilate new information (Mermillod et al., 2013). As a consequence of catastrophic forgetting, performance on previous tasks often drops significantly; in the worst case, previously learned information is completely overwritten by the new one (Parisi et al., 2019). Humans, however, excel at CL by incrementally acquiring, consolidating, and transferring knowledge across tasks (Bremner et al., 2012). Although there is gracious forgetting in humans, learning new information rarely causes catastrophic forgetting of consolidated knowledge (French, 1999).

The in-context learning capabilities of LLMs like GPT-3 allow annotators to customize an LLM to their specific tasks with a small number of examples. However, users tend to include only the most obvious patterns when crafting examples, resulting in underspecified in-context functions that fall short on unseen cases. Further, it is hard to know when "enough" examples have been included even for known patterns. In this work, we present ScatterShot, an interactive system for building high-quality demonstration sets for in-context learning. ScatterShot iteratively slices unlabeled data into task-specific patterns, samples informative inputs from underexplored or not-yet-saturated slices in an active learning manner, and helps users label more efficiently with the help of an LLM and the current example set. In simulation studies on two text perturbation scenarios, ScatterShot sampling improves the resulting few-shot functions by 4-5 percentage points over random sampling, with less variance as more examples are added. In a user study, ScatterShot greatly helps users in covering different patterns in the input space and labeling in-context examples more efficiently, resulting in better in-context learning and less user effort.

Knowledge tracing (KT) is the problem of predicting students' future performance based on their historical interactions with intelligent tutoring systems. Recent studies have applied multiple types of deep neural networks to solve the KT problem. However, there are two important factors in real-world educational data that are not well represented. First, most existing works augment input representations with the co-occurrence matrix of questions and knowledge components\footnote{\label{ft:kc}A KC is a generalization of everyday terms like concept, principle, fact, or skill.} (KCs) but fail to explicitly integrate such intrinsic relations into the final response prediction task. Second, the individualized historical performance of students has not been well captured. In this paper, we proposed \emph{AT-DKT} to improve the prediction performance of the original deep knowledge tracing model with two auxiliary learning tasks, i.e., \emph{question tagging (QT) prediction task} and \emph{individualized prior knowledge (IK) prediction task}. Specifically, the QT task helps learn better question representations by predicting whether questions contain specific KCs. The IK task captures students' global historical performance by progressively predicting student-level prior knowledge that is hidden in students' historical learning interactions. We conduct comprehensive experiments on three real-world educational datasets and compare the proposed approach to both deep sequential KT models and non-sequential models. Experimental results show that \emph{AT-DKT} outperforms all sequential models with more than 0.9\% improvements of AUC for all datasets, and is almost the second best compared to non-sequential models. Furthermore, we conduct both ablation studies and quantitative analysis to show the effectiveness of auxiliary tasks and the superior prediction outcomes of \emph{AT-DKT}.

As a result, they are often brittle and unable to adapt to new domains, can treat individuals or subgroups unfairly, and have limited ability to explain their actions or recommendations [197, 235] reducing the trust of human users [118]. Following this, a new area of research, trustworthy AI, has recently received much attention from several policymakers and other regulatory organizations. The resulting guidelines (e.g., [184, 186, 187]), introduced to increase trust in AI systems, make developing trustworthy AI not only a technical (research) and social endeavor but also an organizational and (legal) obligational requirement. In this paper, we set out to demonstrate, through an extensive survey, that causal modeling and reasoning is an emerging and very useful tool for enabling current AI systems to become trustworthy. Causality is the science of reasoning about causes and effects. Cause-and-effect relationships are central to how we make sense of the world around us, how we act upon it, and how we respond to changes in our environment. In AI, research in causality was pioneered by the Turing award winner Judea Pearl long back in his 1995 seminal paper [194]. Since then, many researchers have contributed to the development of a solid mathematical basis for causality; see, for example, the books [79, 196, 201], the survey [90] and seminal papers [197, 235].

Humans excel at lifelong learning, as the brain has evolved to be robust to distribution shifts and noise in our ever-changing environment. Deep neural networks (DNNs), however, exhibit catastrophic forgetting and the learned representations drift drastically as they encounter a new task. To this end, we propose ESMER which employs a principled mechanism to modulate error sensitivity in a dual-memory rehearsalbased system. Concretely, it maintains a memory of past errors and uses it to modify the learning dynamics so that the model learns more from small consistent errors compared to large sudden errors. We also propose Error-Sensitive Reservoir Sampling to maintain episodic memory, which leverages the error history to pre-select low-loss samples as candidates for the buffer, which are better suited for retaining information. Empirical results show that ESMER effectively reduces forgetting and abrupt drift in representations at the task boundary by gradually adapting to the new task while consolidating knowledge. Remarkably, it also enables the model to learn under high levels of label noise, which is ubiquitous in real-world data streams. The human brain has evolved to engage with and learn from an ever-changing and noisy environment, enabling humans to excel at lifelong learning. This requires it to be robust to varying degrees of distribution shifts and noise to acquire, consolidate, and transfer knowledge under uncertainty. DNNs, on the other hand, are inherently designed for batch learning from a static data distribution and therefore exhibit catastrophic forgetting (McCloskey & Cohen, 1989) of previous tasks when learning tasks sequentially from a continuous stream of data. The significant gap between the lifelong learning capabilities of humans and DNNs suggests that the brain relies on fundamentally different error-based learning mechanisms.

With X-ray free-electron lasers (XFELs), it is possible to determine the three-dimensional structure of noncrystalline nanoscale particles using X-ray single-particle imaging (SPI) techniques at room temperature. Classifying SPI scattering patterns, or "speckles", to extract single hits that are needed for real-time vetoing and three-dimensional reconstruction poses a challenge for high data rate facilities like European XFEL and LCLS-II-HE. Here, we introduce SpeckleNN, a unified embedding model for real-time speckle pattern classification with limited labeled examples that can scale linearly with dataset size. Trained with twin neural networks, SpeckleNN maps speckle patterns to a unified embedding vector space, where similarity is measured by Euclidean distance. We highlight its few-shot classification capability on new never-seen samples and its robust performance despite only tens of labels per classification category even in the presence of substantial missing detector areas. Without the need for excessive manual labeling or even a full detector image, our classification method offers a great solution for real-time high-throughput SPI experiments.

Deep learning has contributed greatly to many successes in artificial intelligence in recent years. Today, it is possible to train models that have thousands of layers and hundreds of billions of parameters. Large-scale deep models have achieved great success, but the enormous computational complexity and gigantic storage requirements make it extremely difficult to implement them in real-time applications. On the other hand, the size of the dataset is still a real problem in many domains. Data are often missing, too expensive, or impossible to obtain for other reasons. Ensemble learning is partially a solution to the problem of small datasets and overfitting. However, ensemble learning in its basic version is associated with a linear increase in computational complexity. We analyzed the impact of the ensemble decision-fusion mechanism and checked various methods of sharing the decisions including voting algorithms. We used the modified knowledge distillation framework as a decision-fusion mechanism which allows in addition compressing of the entire ensemble model into a weight space of a single model. We showed that knowledge distillation can aggregate knowledge from multiple teachers in only one student model and, with the same computational complexity, obtain a better-performing model compared to a model trained in the standard manner. We have developed our own method for mimicking the responses of all teachers at the same time, simultaneously. We tested these solutions on several benchmark datasets. In the end, we presented a wide application use of the efficient multi-teacher knowledge distillation framework. In the first example, we used knowledge distillation to develop models that could automate corrosion detection on aircraft fuselage. The second example describes detection of smoke on observation cameras in order to counteract wildfires in forests.

This paper presents Deep-PANTHER, a learning-based perception-aware trajectory planner for unmanned aerial vehicles (UAVs) in dynamic environments. Given the current state of the UAV, and the predicted trajectory and size of the obstacle, Deep-PANTHER generates multiple trajectories to avoid a dynamic obstacle while simultaneously maximizing its presence in the field of view (FOV) of the onboard camera. To obtain a computationally tractable real-time solution, imitation learning is leveraged to train a Deep-PANTHER policy using demonstrations provided by a multimodal optimization-based expert. Extensive simulations show replanning times that are two orders of magnitude faster than the optimization-based expert, while achieving a similar cost. By ensuring that each expert trajectory is assigned to one distinct student trajectory in the loss function, Deep-PANTHER can also capture the multimodality of the problem and achieve a mean squared error (MSE) loss with respect to the expert that is up to 18 times smaller than state-of-the-art (Relaxed) Winner-Takes-All approaches. Deep-PANTHER is also shown to generalize well to obstacle trajectories that differ from the ones used in training.