ML workloads are becoming increasingly popular in the cloud. Good cloud training performance is contingent on efficient parameter exchange among VMs. We find that Collectives, the widely used distributed communication algorithms, cannot perform optimally out of the box due to the hierarchical topology of datacenter networks and multi-tenancy nature of the cloudenvironment.In this paper, we present Cloud Collectives , a prototype that accelerates collectives by reordering theranks of participating VMs such that the communication pattern dictated by the selected collectives operation best exploits the locality in the network.Collectives is non-intrusive, requires no code changes nor rebuild of an existing application, and runs without support from cloud providers. Our preliminary application of Cloud Collectives on allreduce operations in public clouds results in a speedup of up to 3.7x in multiple microbenchmarks and 1.3x in real-world workloads of distributed training of deep neural networks and gradient boosted decision trees using state-of-the-art frameworks.

NLP community is currently investing a lot more research and resources into development of deep learning models than training data. While we have made a lot of progress, it is now clear that our models learn all kinds of spurious patterns, social biases, and annotation artifacts. Algorithmic solutions have so far had limited success. An alternative that is being actively discussed is more careful design of datasets so as to deliver specific signals. This position paper maps out the arguments for and against data curation, and argues that fundamentally the point is moot: curation already is and will be happening, and it is changing the world. The question is only how much thought we want to invest into that process.

Recent explorations of large-scale pre-trained language models (PLMs) such as GPT-3 have revealed the power of PLMs with huge amounts of parameters, setting off a wave of training ever-larger PLMs. However, training a large-scale PLM requires tremendous amounts of computational resources, which is time-consuming and expensive. In addition, existing large-scale PLMs are mainly trained from scratch individually, ignoring the availability of many existing well-trained PLMs. To this end, we explore the question that how can previously trained PLMs benefit training larger PLMs in future. Specifically, we introduce a novel pre-training framework named "knowledge inheritance" (KI), which combines both self-learning and teacher-guided learning to efficiently train larger PLMs. Sufficient experimental results demonstrate the feasibility of our KI framework. We also conduct empirical analyses to explore the effects of teacher PLMs' pre-training settings, including model architecture, pre-training data, etc. Finally, we show that KI can well support lifelong learning and knowledge transfer.

Recent interest in on-orbit servicing and Active Debris Removal (ADR) missions have driven the need for technologies to enable non-cooperative rendezvous manoeuvres. Such manoeuvres put heavy burden on the perception capabilities of a chaser spacecraft. This paper demonstrates Convolutional Neural Networks (CNNs) capable of providing an initial coarse pose estimation of a target from a passive thermal infrared camera feed. Thermal cameras offer a promising alternative to visible cameras, which struggle in low light conditions and are susceptible to overexposure. Often, thermal information on the target is not available a priori; this paper therefore proposes using visible images to train networks. The robustness of the models is demonstrated on two different targets, first on synthetic data, and then in a laboratory environment for a realistic scenario that might be faced during an ADR mission. Given that there is much concern over the use of CNN in critical applications due to their black box nature, we use innovative techniques to explain what is important to our network and fault conditions.

Data augmentation is an effective way to improve the performance of many neural text generation models. However, current data augmentation methods need to define or choose proper data mapping functions that map the original samples into the augmented samples. In this work, we derive an objective to formulate the problem of data augmentation on text generation tasks without any use of augmented data constructed by specific mapping functions. Our proposed objective can be efficiently optimized and applied to popular loss functions on text generation tasks with a convergence rate guarantee. Experiments on five datasets of two text generation tasks show that our approach can approximate or even surpass popular data augmentation methods.

Data used in deep learning is notoriously problematic. For example, data are usually combined from diverse sources, rarely cleaned and vetted thoroughly, and sometimes corrupted on purpose. Intentional corruption that targets the weak spots of algorithms has been studied extensively under the label of "adversarial attacks." In contrast, the arguably much more common case of corruption that reflects the limited quality of data has been studied much less. Such "random" corruptions are due to measurement errors, unreliable sources, convenience sampling, and so forth. These kinds of corruption are common in deep learning, because data are rarely collected according to strict protocols -- in strong contrast to the formalized data collection in some parts of classical statistics. This paper concerns such corruption. We introduce an approach motivated by very recent insights into median-of-means and Le Cam's principle, we show that the approach can be readily implemented, and we demonstrate that it performs very well in practice. In conclusion, we believe that our approach is a very promising alternative to standard parameter training based on least-squares and cross-entropy loss.

Attribution methods provide an insight into the decision-making process of machine learning models, especially deep neural networks, by assigning contribution scores to each individual feature. However, the attribution problem has not been well-defined, which lacks a unified guideline to the contribution assignment process. Furthermore, existing attribution methods often built upon various empirical intuitions and heuristics. There still lacks a general theoretical framework that not only can offer a good description of the attribution problem, but also can be applied to unifying and revisiting existing attribution methods. To bridge the gap, in this paper, we propose a Taylor attribution framework, which models the attribution problem as how to decide individual payoffs in a coalition. Then, we reformulate fourteen mainstream attribution methods into the Taylor framework and analyze these attribution methods in terms of rationale, fidelity, and limitation in the framework. Moreover, we establish three principles for a good attribution in the Taylor attribution framework, i.e., low approximation error, correct Taylor contribution assignment, and unbiased baseline selection. Finally, we empirically validate the Taylor reformulations and reveal a positive correlation between the attribution performance and the number of principles followed by the attribution method via benchmarking on real-world datasets.

In this paper, we propose a new framework to detect adversarial examples motivated by the observations that random components can improve the smoothness of predictors and make it easier to simulate output distribution of deep neural network. With these observations, we propose a novel Bayesian adversarial example detector, short for BATer, to improve the performance of adversarial example detection. In specific, we study the distributional difference of hidden layer output between natural and adversarial examples, and propose to use the randomness of Bayesian neural network (BNN) to simulate hidden layer output distribution and leverage the distribution dispersion to detect adversarial examples. The advantage of BNN is that the output is stochastic while neural networks without random components do not have such characteristics. Empirical results on several benchmark datasets against popular attacks show that the proposed BATer outperforms the state-of-the-art detectors in adversarial example detection.

" Star Wars: A New Hope ", the 1977 classic directed by George Lucas, was, for many of us, the first story that brought us closer and amazed by showing us a hologram . " Help me Obi-Wan Kenobi, you are my last hope, " said the hologram of Princess Leia projected by R2D2 in the installment of " Star Wars: Episode IV - A New Hope. The truth is that this innovative technology has transcended screens and science fiction and is currently being developed successfully thanks to advances in Artificial Intelligence . Until today, virtual reality headsets have not brought down television screens and computers as a reference device for watching videos. A reason? virtual reality can make users feel bad.

Deep Learning is really just a subset of Machine Learning that has garnered significant attention recently due to its stellar performance across many tasks as we've listed above. That begs the question -- What is Machine Learning? And how is Machine Learning any different from "traditional algorithms"? If you've taken a class on algorithms, the standard metaphor for an algorithm is a recipe. An algorithm is a series of steps, that when performed in a specific order, produced your desired output. A "cooking algorithm" for a robot to make bread might go something like this: And so on… (I'm not a bread expert, so I shan't write a recipe for bread here) In Machine Learning, we don't specify the algorithm the way we did above.