MOTIVATION: Extensive research has uncovered the critical role of the human gut microbiome in various aspects of health, including metabolism, nutrition, physiology, and immune function. Fecal microbiota is often used as a proxy for understanding the gut microbiome, but it represents an aggregate view, overlooking spatial variations across different gastrointestinal (GI) locations. Emerging studies with spatial microbiome data collected from specific GI regions offer a unique opportunity to better understand the spatial composition of the stool microbiome. RESULTS: We introduce Micro-DeMix, a mixture beta-multinomial model that deconvolutes the fecal microbiome at the compositional level by integrating stool samples with spatial microbiome data. Micro-DeMix facilitates the comparison of microbial compositions across different GI regions within the stool microbiome through a hypothesis-testing framework. We demonstrate the effectiveness and efficiency of Micro-DeMix using multiple simulated datasets and the inflammatory bowel disease data from the NIH Integrative Human Microbiome Project. AVAILABILITY AND IMPLEMENTATION: The R package is available at https://github.com/liuruoqian/MicroDemix.

Rosanne Liu, Dan Garrette, Chitwan Saharia, William Chan, Adam Roberts, Sharan Narang, Irina Blok, Rj Mical, Mohammad Norouzi, Noah Constant. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Open-vocabulary models are a promising new paradigm for image classification. Unlike traditional classification models, open-vocabulary models classify among any arbitrary set of categories specified with natural language during inference. This natural language, called "prompts", typically consists of a set of hand-written templates (e.g., "a photo of a {}") which are completed with each of the category names. This work introduces a simple method to generate higher accuracy prompts, without relying on any explicit knowledge of the task domain and with far fewer hand-constructed sentences. To achieve this, we combine open-vocabulary models with large language models (LLMs) to create Customized Prompts via Language models (CuPL, pronounced "couple"). In particular, we leverage the knowledge contained in LLMs in order to generate many descriptive sentences that contain important discriminating characteristics of the image categories. This allows the model to place a greater importance on these regions in the image when making predictions. We find that this straightforward and general approach improves accuracy on a range of zero-shot image classification benchmarks, including over one percentage point gain on ImageNet. Finally, this simple baseline requires no additional training and remains completely zero-shot. Code available at https://github.com/sarahpratt/CuPL.

In this paper, we introduce a novel form of value function, $Q(s, s’)$, that expresses the utility of transitioning from a state $s$ to a neighboring state $s’$ and then acting optimally thereafter. In order to derive an optimal policy, we develop a forward dynamics model that learns to make next-state predictions that maximize this value. This formulation decouples actions from values while still learning off-policy. We highlight the benefits of this approach in terms of value function transfer, learning within redundant action spaces, and learning off-policy from state observations generated by sub-optimal or completely random policies. Code and videos are available at http://sites.google.com/view/qss-paper.

We present the Supermasks in Superposition (SupSup) model, capable of sequentially learning thousands of tasks without catastrophic forgetting. Our approach uses a randomly initialized, fixed base network and for each task finds a subnetwork (supermask) that achieves good performance. If task identity is given at test time, the correct subnetwork can be retrieved with minimal memory usage. If not provided, SupSup can infer the task using gradient-based optimization to find a linear superposition of learned supermasks which minimizes the output entropy. In practice we find that a single gradient step is often sufficient to identify the correct mask, even among 2500 tasks. We also showcase two promising extensions. First, SupSup models can be trained entirely without task identity information, as they may detect when they are uncertain about new data and allocate an additional supermask for the new training distribution. Finally the entire, growing set of supermasks can be stored in a constant-sized reservoir by implicitly storing them as attractors in a fixed-sized Hopfield network.

Much human and computational effort has aimed to improve how deep reinforcement learning (DRL) algorithms perform on benchmarks such as the Atari Learning Environment. Comparatively less effort has focused on understanding what has been learned by such methods, and investigating and comparing the representations learned by different families of DRL algorithms. Sources of friction include the onerous computational requirements, and general logistical and architectural complications for running DRL algorithms at scale. We lessen this friction, by (1) training several algorithms at scale and releasing trained models, (2) integrating with a previous DRL model release, and (3) releasing code that makes it easy for anyone to load, visualize, and analyze such models. This paper introduces the Atari Zoo framework, which contains models trained across benchmark Atari games, in an easy-to-use format, as well as code that implements common modes of analysis and connects such models to a popular neural network visualization library. Further, to demonstrate the potential of this dataset and software package, we show initial quantitative and qualitative comparisons between the performance and representations of several DRL algorithms, highlighting interesting and previously unknown distinctions between them.

Neural networks enjoy widespread use, but many aspects of their training, representation, and operation are poorly understood. In particular, our view into the training process is limited, with a single scalar loss being the most common viewport into this high-dimensional, dynamic process. We propose a new window into training called Loss Change Allocation (LCA), in which credit for changes to the network loss is conservatively partitioned to the parameters. This measurement is accomplished by decomposing the components of an approximate path integral along the training trajectory using a Runge-Kutta integrator. This rich view shows which parameters are responsible for decreasing or increasing the loss during training, or which parameters "help" or "hurt" the network's learning, respectively. LCA may be summed over training iterations and/or over neurons, channels, or layers for increasingly coarse views. This new measurement device produces several insights into training. (1) We find that barely over 50% of parameters help during any given iteration. (2) Some entire layers hurt overall, moving on average against the training gradient, a phenomenon we hypothesize may be due to phase lag in an oscillatory training process. (3) Finally, increments in learning proceed in a synchronized manner across layers, often peaking on identical iterations.

The recent "Lottery Ticket Hypothesis" paper by Frankle & Carbin showed that a simple approach to creating sparse networks (keep the large weights) results in models that are trainable from scratch, but only when starting from the same initial weights. The performance of these networks often exceeds the performance of the non-sparse base model, but for reasons that were not well understood. In this paper we study the three critical components of the Lottery Ticket (LT) algorithm, showing that each may be varied significantly without impacting the overall results. Ablating these factors leads to new insights for why LT networks perform as well as they do. We show why setting weights to zero is important, how signs are all you need to make the re-initialized network train, and why masking behaves like training. Finally, we discover the existence of Supermasks, or masks that can be applied to an untrained, randomly initialized network to produce a model with performance far better than chance (86% on MNIST, 41% on CIFAR-10).

The simple, elegant approach of training convolutional neural networks (CNNs) directly from RGB pixels has enjoyed overwhelming empirical success. But can more performance be squeezed out of networks by using different input representations? In this paper we propose and explore a simple idea: train CNNs directly on the blockwise discrete cosine transform (DCT) coefficients computed and available in the middle of the JPEG codec. Intuitively, when processing JPEG images using CNNs, it seems unnecessary to decompress a blockwise frequency representation to an expanded pixel representation, shuffle it from CPU to GPU, and then process it with a CNN that will learn something similar to a transform back to frequency representation in its first layers. Why not skip both steps and feed the frequency domain into the network directly? In this paper we modify \libjpeg to produce DCT coefficients directly, modify a ResNet-50 network to accommodate the differently sized and strided input, and evaluate performance on ImageNet. We find networks that are both faster and more accurate, as well as networks with about the same accuracy but 1.77x faster than ResNet-50.

Few ideas have enjoyed as large an impact on deep learning as convolution. For any problem involving pixels or spatial representations, common intuition holds that convolutional neural networks may be appropriate. In this paper we show a striking counterexample to this intuition via the seemingly trivial coordinate transform problem, which simply requires learning a mapping between coordinates in (x,y) Cartesian space and coordinates in one-hot pixel space. Although convolutional networks would seem appropriate for this task, we show that they fail spectacularly. We demonstrate and carefully analyze the failure first on a toy problem, at which point a simple fix becomes obvious. We call this solution CoordConv, which works by giving convolution access to its own input coordinates through the use of extra coordinate channels. Without sacrificing the computational and parametric efficiency of ordinary convolution, CoordConv allows networks to learn either complete translation invariance or varying degrees of translation dependence, as required by the end task. CoordConv solves the coordinate transform problem with perfect generalization and 150 times faster with 10--100 times fewer parameters than convolution. This stark contrast raises the question: to what extent has this inability of convolution persisted insidiously inside other tasks, subtly hampering performance from within? A complete answer to this question will require further investigation, but we show preliminary evidence that swapping convolution for CoordConv can improve models on a diverse set of tasks. Using CoordConv in a GAN produced less mode collapse as the transform between high-level spatial latents and pixels becomes easier to learn. A Faster R-CNN detection model trained on MNIST detection showed 24% better IOU when using CoordConv, and in the Reinforcement Learning (RL) domain agents playing Atari games benefit significantly from the use of CoordConv layers.

This paper explores the idea of modeling a large image data collection of polycrystal electron patterns, in order to detect insights in understanding materials discovery. There is an emerging interest in applying big data processing, management and modeling methods to scientific images, which often come in a form and with patterns only interpretable to domain experts. While large-scale machine learning approaches have demonstrated certain superiority in analyzing, summarizing, and providing an understandable route to data types like natural images, speeches and texts, scientific images is still a relatively unexplored area. Deep convolutional neural networks, despite their recent triumph in natural image understanding, are still rarely seen adapted to experimental microscopic images, especially in a large scale. To the best of our knowledge, we present the first deep learning solution towards a scientific image indexing problem using a collection of over 300K microscopic images. The result obtained is 54% better than a dictionary lookup method which is state-of-the-art in the materials science society.

Recent progress in big data and computer vision with deep learning models has gained a lot of attention. Deep learning has been performed on tasks such as image classification, object detection, image segmentation, image captioning, visual question and answering, using large collections of annotated images. This calls for more curated large image datasets with clearer descriptions, cleaner contents, and diversified usability. However, the curation and labeling of such datasets can be labor-intensive. In this paper, we present PinterNet, an algorithm for automatic curation and label generation from noisy textual descriptions, and also publish a big image dataset containing over 110K images automatically labeled with their themes. Our dataset is hierarchical in nature, it has high level category information which we refer as verticals with fine-grained thematic labels at lower level. This advocates a new type of hierarchical theme classification problem closer to human cognition and of business value. We provide benchmark performances using deep learning models based on AlexNet architecture with different pre-training schemes for this novel task and new data.

This paper presents the methodologies developed for solving IJCNN 2011's Ford Challenge II problem, where the driver's alertness is to be detected employing physiological, environmental and vehicular data acquired during driving. The solution is based on a thorough four-fold framework consisting of temporal processing, feature creation and extraction, and the training and ensemble of several learning systems, such as neural networks, random forest, support vector machine, trained from diverse features. The selection of input features to a learning machine has always been critique on signal classification. In our approach, the employment of Bayesian network filtered out a set of features and has been proved by the ensemble to be effective. The ensemble technique enhanced the performance of individual systems dramatically. The performance acquired on 30% of the test samples reached an accuracy of 78.34%. These results are significant for a real-world vehicular problem and we are quite confident this solution will become one of the top ones on the competition test data.

Time series data classification is important in many applications. Learning temporal knowledge in time series data is challenging. In this paper we propose a novel machine learning algorithm, Feature Ensemble (FE), to learn effective subsequences of signal features distributed over time series data streams. Both the FE learning and the FE classification have been applied to an application problem. Our empirical results strongly suggest that FE learning is an effective technique for time series data classification.