Agriculture is the foundation of human civilization. However, the rapid increase and aging of the global population pose challenges on this cornerstone by demanding more healthy and fresh food. Internet of Things (IoT) technology makes modern autonomous greenhouse a viable and reliable engine of food production. However, the educated and skilled labor capable of overseeing high-tech greenhouses is scarce. Artificial intelligence (AI) and cloud computing technologies are promising solutions for precision control and high-efficiency production in such controlled environments. In this paper, we propose a smart agriculture solution, namely iGrow: (1) we use IoT and cloud computing technologies to measure, collect, and manage growing data, to support iteration of our decision-making AI module, which consists of an incremental model and an optimization algorithm; (2) we propose a three-stage incremental model based on accumulating data, enabling growers/central computers to schedule control strategies conveniently and at low cost; (3) we propose a model-based iterative optimization algorithm, which can dynamically optimize the greenhouse control strategy in real-time production. In the simulated experiment, evaluation results show the accuracy of our incremental model is comparable to an advanced tomato simulator, while our optimization algorithms can beat the champion of the 2nd Autonomous Greenhouse Challenge. Compelling results from the A/B test in real greenhouses demonstrate that our solution significantly increases production (commercially sellable fruits) (+ 10.15%) and net profit (+ 87.07%) with statistical significance compared to planting experts.

Perhaps, the modern definition of AI - as the ensemble of computer systems empowered with the ability to learn from data through statistical techniques - can be dated back to 1959. Machine Learning (ML), a subclass of AI, is a discipline that aims at studying algorithms that are able to learn from experience and data to perform tasks without following explicit instructions. Often, these algorithms are based on a computational model that belongs to differentiable programming techniques, called Neural Networks (NNs). The success of such algorithms resides in their ability to learn to achieve a specific goal [95, 121], i.e., they learn to discover hidden patterns and relations among data to fulfill the task at hand [89, 120]. Mathematically, NNs are made of a sequence of transformations, called layers, composed of affine operators and elementwise nonlinearities. Then, the goal of learning is to modify the transformations' parameters to fulfill a task successfully. Whenever a model accounts for more than a couple of such layers, it is called a Deep Learning (DL) model or a Deep Neural Network (DNN). Thanks to their enormous representation power and the development of new technologies and training algorithms, DL models obtained astonishing results in the last two decades, achieving superhuman performance on certain tasks [177].

We show that a popular self-supervised learning method, InfoNCE, is a special case of a new family of unsupervised learning methods, the self-supervised variational autoencoder (SSVAE). SSVAEs circumvent the usual VAE requirement to reconstruct the data by using a carefully chosen implicit decoder. The InfoNCE objective was motivated as a simplified parametric mutual information estimator. Under one choice of prior, the SSVAE objective (i.e. the ELBO) is exactly equal to the mutual information (up to constants). Under an alternative choice of prior, the SSVAE objective is exactly equal to the simplified parametric mutual information estimator used in InfoNCE (up to constants). Importantly, the use of simplified parametric mutual information estimators is believed to be critical to obtain good high-level representations, and the SSVAE framework naturally provides a principled justification for using prior information to choose these estimators.

Deep predictive models often make use of spurious correlations between the label and the covariates that differ between training and test distributions. In many classification tasks, spurious correlations are induced by a changing relationship between the label and some nuisance variables correlated with the covariates. For example, in classifying animals in natural images, the background, which is the nuisance, can predict the type of animal. This nuisance-label relationship does not always hold. We formalize a family of distributions that only differ in the nuisance-label relationship and introduce a distribution where this relationship is broken called the nuisance-randomized distribution. We introduce a set of predictive models built from the nuisance-randomized distribution with representations, that when conditioned on, do not correlate the label and the nuisance. For models in this set, we lower bound the performance for any member of the family with the mutual information between the representation and the label under the nuisance-randomized distribution. To build predictive models that maximize the performance lower bound, we develop Nuisance-Randomized Distillation (NURD). We evaluate NURD on a synthetic example, colored-MNIST, and classifying chest X-rays. When using non-lung patches as the nuisance in classifying chest X-rays, NURD produces models that predict pneumonia under strong spurious correlations.

Hyperspectral (HS) images are characterized by approximately contiguous spectral information, enabling the fine identification of materials by capturing subtle spectral discrepancies. Owing to their excellent locally contextual modeling ability, convolutional neural networks (CNNs) have been proven to be a powerful feature extractor in HS image classification. However, CNNs fail to mine and represent the sequence attributes of spectral signatures well due to the limitations of their inherent network backbone. To solve this issue, we rethink HS image classification from a sequential perspective with transformers, and propose a novel backbone network called \ul{SpectralFormer}. Beyond band-wise representations in classic transformers, SpectralFormer is capable of learning spectrally local sequence information from neighboring bands of HS images, yielding group-wise spectral embeddings. More significantly, to reduce the possibility of losing valuable information in the layer-wise propagation process, we devise a cross-layer skip connection to convey memory-like components from shallow to deep layers by adaptively learning to fuse "soft" residuals across layers. It is worth noting that the proposed SpectralFormer is a highly flexible backbone network, which can be applicable to both pixel- and patch-wise inputs. We evaluate the classification performance of the proposed SpectralFormer on three HS datasets by conducting extensive experiments, showing the superiority over classic transformers and achieving a significant improvement in comparison with state-of-the-art backbone networks. The codes of this work will be available at \url{https://sites.google.com/view/danfeng-hong} for the sake of reproducibility.

Electronic health records (EHRs), digital collections of patient healthcare events and observations, are ubiquitous in medicine and critical to healthcare delivery, operations, and research. Despite this central role, EHRs are notoriously difficult to process automatically. Well over half of the information stored within EHRs is in the form of unstructured text (e.g. provider notes, operation reports) and remains largely untapped for secondary use. Recently, however, newer neural network and deep learning approaches to Natural Language Processing (NLP) have made considerable advances, outperforming traditional statistical and rule-based systems on a variety of tasks. In this survey paper, we summarize current neural NLP methods for EHR applications. We focus on a broad scope of tasks, namely, classification and prediction, word embeddings, extraction, generation, and other topics such as question answering, phenotyping, knowledge graphs, medical dialogue, multilinguality, interpretability, etc.

Building vehicles capable of operating without human supervision requires the determination of the agent's pose. Visual Odometry (VO) algorithms estimate the egomotion using only visual changes from the input images. The most recent VO methods implement deep-learning techniques using convolutional neural networks (CNN) extensively, which add a substantial cost when dealing with high-resolution images. Furthermore, in VO tasks, more input data does not mean a better prediction; on the contrary, the architecture may filter out useless information. Therefore, the implementation of computationally efficient and lightweight architectures is essential. In this work, we propose the RAM-VO, an extension of the Recurrent Attention Model (RAM) for visual odometry tasks. RAM-VO improves the visual and temporal representation of information and implements the Proximal Policy Optimization (PPO) algorithm to learn robust policies. The results indicate that RAM-VO can perform regressions with six degrees of freedom from monocular input images using approximately 3 million parameters. In addition, experiments on the KITTI dataset demonstrate that RAM-VO achieves competitive results using only 5.7% of the available visual information.

When observing task demonstrations, human apprentices are able to identify whether a given task is executed correctly long before they gain expertise in actually performing that task. Prior research into learning from demonstrations (LfD) has failed to capture this notion of the acceptability of a task's execution; meanwhile, temporal logics provide a flexible language for expressing task specifications. Inspired by this, we present Bayesian specification inference, a probabilistic model for inferring task specification as a temporal logic formula. We incorporate methods from probabilistic programming to define our priors, along with a domain-independent likelihood function to enable sampling-based inference. We demonstrate the efficacy of our model for inferring specifications, with over 90% similarity observed between the inferred specification and the ground truth, both within a synthetic domain and during a real-world table setting task.

The goal of Question Answering over Knowledge Graphs (KGQA) is to find answers for natural language questions over a knowledge graph. Recent KGQA approaches adopt a neural machine translation (NMT) approach, where the natural language question is translated into a structured query language. However, NMT suffers from the out-of-vocabulary problem, where terms in a question may not have been seen during training, impeding their translation. This issue is particularly problematic for the millions of entities that large knowledge graphs describe. We rather propose a KGQA approach that delegates the processing of entities to entity linking (EL) systems. NMT is then used to create a query template with placeholders that are filled by entities identified in an EL phase. Slot filling is used to decide which entity fills which placeholder. Experiments for QA over Wikidata show that our approach outperforms pure NMT: while there remains a strong dependence on having seen similar query templates during training, errors relating to entities are greatly reduced.

Human reasoning can often be understood as an interplay between two systems: the intuitive and associative ("System 1") and the deliberative and logical ("System 2"). Neural sequence models -- which have been increasingly successful at performing complex, structured tasks -- exhibit the advantages and failure modes of System 1: they are fast and learn patterns from data, but are often inconsistent and incoherent. In this work, we seek a lightweight, training-free means of improving existing System 1-like sequence models by adding System 2-inspired logical reasoning. We explore several variations on this theme in which candidate generations from a neural sequence model are examined for logical consistency by a symbolic reasoning module, which can either accept or reject the generations. Our approach uses neural inference to mediate between the neural System 1 and the logical System 2. Results in robust story generation and grounded instruction-following show that this approach can increase the coherence and accuracy of neurally-based generations.