We present a new zero-shot dense retrieval (ZeroDR) method, COCO-DR, to improve the generalization ability of dense retrieval by combating the distribution shifts between source training tasks and target scenarios. To mitigate the impact of document differences, COCO-DR continues pretraining the language model on the target corpora to adapt the model to target distributions via COtinuous COtrastive learning. To prepare for unseen target queries, COCO-DR leverages implicit Distributionally Robust Optimization (iDRO) to reweight samples from different source query clusters for improving model robustness over rare queries during fine-tuning. COCO-DR achieves superior average performance on BEIR, the zero-shot retrieval benchmark. At BERT Base scale, COCO-DR Base outperforms other ZeroDR models with 60x larger size. At BERT Large scale, COCO-DR Large outperforms the giant GPT-3 embedding model which has 500x more parameters. Our analysis show the correlation between COCO-DR's effectiveness in combating distribution shifts and improving zero-shot accuracy. Our code and model can be found at \url{https://github.com/OpenMatch/COCO-DR}.

Multivariate time series forecasting has long received significant attention in real-world applications, such as energy consumption and traffic prediction. While recent methods demonstrate good forecasting abilities, they have three fundamental limitations. (i) Discrete neural architectures: Interlacing individually parameterized spatial and temporal blocks to encode rich underlying patterns leads to discontinuous latent state trajectories and higher forecasting numerical errors. (ii) High complexity: Discrete approaches complicate models with dedicated designs and redundant parameters, leading to higher computational and memory overheads. (iii) Reliance on graph priors: Relying on predefined static graph structures limits their effectiveness and practicability in real-world applications. In this paper, we address all the above limitations by proposing a continuous model to forecast $\textbf{M}$ultivariate $\textbf{T}$ime series with dynamic $\textbf{G}$raph neural $\textbf{O}$rdinary $\textbf{D}$ifferential $\textbf{E}$quations ($\texttt{MTGODE}$). Specifically, we first abstract multivariate time series into dynamic graphs with time-evolving node features and unknown graph structures. Then, we design and solve a neural ODE to complement missing graph topologies and unify both spatial and temporal message passing, allowing deeper graph propagation and fine-grained temporal information aggregation to characterize stable and precise latent spatial-temporal dynamics. Our experiments demonstrate the superiorities of $\texttt{MTGODE}$ from various perspectives on five time series benchmark datasets.

The 21 cm spectral line emission of atomic neutral hydrogen (HI) is one of the primary wavelengths observed in radio astronomy. However, the signal is intrinsically faint and the HI content of galaxies depends on the cosmic environment, requiring large survey volumes and survey depth to investigate the HI Universe. As the amount of data coming from these surveys continues to increase with technological improvements, so does the need for automatic techniques for identifying and characterising HI sources while considering the tradeoff between completeness and purity. This study aimed to find the optimal pipeline for finding and masking the most sources with the best mask quality and the fewest artefacts in 3D neutral hydrogen cubes. Various existing methods were explored in an attempt to create a pipeline to optimally identify and mask the sources in 3D neutral hydrogen 21 cm spectral line data cubes. Two traditional source-finding methods were tested, SoFiA and MTObjects, as well as a new supervised deep learning approach, in which a 3D convolutional neural network architecture, known as V-Net was used. These three source-finding methods were further improved by adding a classical machine learning classifier as a post-processing step to remove false positive detections. The pipelines were tested on HI data cubes from the Westerbork Synthesis Radio Telescope with additional inserted mock galaxies. SoFiA combined with a random forest classifier provided the best results, with the V-Net-random forest combination a close second. We suspect this is due to the fact that there are many more mock sources in the training set than real sources. There is, therefore, room to improve the quality of the V-Net network with better-labelled data such that it can potentially outperform SoFiA.

Variational Auto-Encoder (VAE) has been widely adopted in text generation. Among many variants, recurrent VAE learns token-wise latent variables with each conditioned on the preceding ones, which captures sequential variability better in the era of RNN. However, it is unclear how to incorporate such recurrent dynamics into the recently dominant Transformer due to its parallelism. In this work, we propose TRACE, a Transformer-based recurrent VAE structure. TRACE imposes recurrence on segment-wise latent variables with arbitrarily separated text segments and constructs the posterior distribution with residual parameterization. Besides, we design an acceleration method by approximating idempotent matrices, which allows parallelism while maintaining the conditional dependence of latent variables. We demonstrate that TRACE could enhance the entanglement of each segment and preceding latent variables and deduce a non-zero lower bound of the KL term, providing a theoretical guarantee of generation diversity. Experiments on two unconditional and one conditional generation tasks show that TRACE achieves significantly improved diversity while maintaining satisfactory generation quality.

Existing research generally treats Chinese character as a minimum unit for representation. However, such Chinese character representation will suffer two bottlenecks: 1) Learning bottleneck, the learning cannot benefit from its rich internal features (e.g., radicals and strokes); and 2) Parameter bottleneck, each individual character has to be represented by a unique vector. In this paper, we introduce a novel representation method for Chinese characters to break the bottlenecks, namely StrokeNet, which represents a Chinese character by a Latinized stroke sequence (e.g., "ao1 (concave)" to "ajaie" and "tu1 (convex)" to "aeaqe"). Specifically, StrokeNet maps each stroke to a specific Latin character, thus allowing similar Chinese characters to have similar Latin representations. With the introduction of StrokeNet to neural machine translation (NMT), many powerful but not applicable techniques to non-Latin languages (e.g., shared subword vocabulary learning and ciphertext-based data augmentation) can now be perfectly implemented. Experiments on the widely-used NIST Chinese-English, WMT17 Chinese-English and IWSLT17 Japanese-English NMT tasks show that StrokeNet can provide a significant performance boost over the strong baselines with fewer model parameters, achieving 26.5 BLEU on the WMT17 Chinese-English task which is better than any previously reported results without using monolingual data. Code and scripts are freely available at https://github.com/zjwang21/StrokeNet.

Attack Ensemble (AE), which combines multiple attacks together, provides a reliable way to evaluate adversarial robustness. In practice, AEs are often constructed and tuned by human experts, which however tends to be sub-optimal and time-consuming. In this work, we present AutoAE, a conceptually simple approach for automatically constructing AEs. In brief, AutoAE repeatedly adds the attack and its iteration steps to the ensemble that maximizes ensemble improvement per additional iteration consumed. We show theoretically that AutoAE yields AEs provably within a constant factor of the optimal for a given defense. We then use AutoAE to construct two AEs for $l_{\infty}$ and $l_2$ attacks, and apply them without any tuning or adaptation to 45 top adversarial defenses on the RobustBench leaderboard. In all except one cases we achieve equal or better (often the latter) robustness evaluation than existing AEs, and notably, in 29 cases we achieve better robustness evaluation than the best known one. Such performance of AutoAE shows itself as a reliable evaluation protocol for adversarial robustness, which further indicates the huge potential of automatic AE construction. Code is available at \url{https://github.com/LeegerPENG/AutoAE}.

Common Deep Metric Learning (DML) datasets specify only one notion of similarity, e.g., two images in the Cars196 dataset are deemed similar if they show the same car model. We argue that depending on the application, users of image retrieval systems have different and changing similarity notions that should be incorporated as easily as possible. Therefore, we present Language-Guided Zero-Shot Deep Metric Learning (LanZ-DML) as a new DML setting in which users control the properties that should be important for image representations without training data by only using natural language. To this end, we propose InDiReCT (Image representations using Dimensionality Reduction on CLIP embedded Texts), a model for LanZ-DML on images that exclusively uses a few text prompts for training. InDiReCT utilizes CLIP as a fixed feature extractor for images and texts and transfers the variation in text prompt embeddings to the image embedding space. Extensive experiments on five datasets and overall thirteen similarity notions show that, despite not seeing any images during training, InDiReCT performs better than strong baselines and approaches the performance of fully-supervised models. An analysis reveals that InDiReCT learns to focus on regions of the image that correlate with the desired similarity notion, which makes it a fast to train and easy to use method to create custom embedding spaces only using natural language.

Cyber deception is emerging as a promising approach to defending networks and systems against attackers and data thieves. However, despite being relatively cheap to deploy, the generation of realistic content at scale is very costly, due to the fact that rich, interactive deceptive technologies are largely hand-crafted. With recent improvements in Machine Learning, we now have the opportunity to bring scale and automation to the creation of realistic and enticing simulated content. In this work, we propose a framework to automate the generation of email and instant messaging-style group communications at scale. Such messaging platforms within organisations contain a lot of valuable information inside private communications and document attachments, making them an enticing target for an adversary. We address two key aspects of simulating this type of system: modelling when and with whom participants communicate, and generating topical, multi-party text to populate simulated conversation threads. We present the LogNormMix-Net Temporal Point Process as an approach to the first of these, building upon the intensity-free modeling approach of Shchur et al. to create a generative model for unicast and multi-cast communications. We demonstrate the use of fine-tuned, pre-trained language models to generate convincing multi-party conversation threads. A live email server is simulated by uniting our LogNormMix-Net TPP (to generate the communication timestamp, sender and recipients) with the language model, which generates the contents of the multi-party email threads. We evaluate the generated content with respect to a number of realism-based properties, that encourage a model to learn to generate content that will engage the attention of an adversary to achieve a deception outcome.

Large sequence to sequence models for tasks such as Neural Machine Translation (NMT) are usually trained over hundreds of millions of samples. However, training is just the origin of a model's life-cycle. Real-world deployments of models require further behavioral adaptations as new requirements emerge or shortcomings become known. Typically, in the space of model behaviors, behavior deletion requests are addressed through model retrainings whereas model finetuning is done to address behavior addition requests, both procedures being instances of data-based model intervention. In this work, we present a preliminary study investigating rank-one editing as a direct intervention method for behavior deletion requests in encoder-decoder transformer models. We propose four editing tasks for NMT and show that the proposed editing algorithm achieves high efficacy, while requiring only a single instance of positive example to fix an erroneous (negative) model behavior.

Prompt-based methods have been successfully applied in sentence-level few-shot learning tasks, mostly owing to the sophisticated design of templates and label words. However, when applied to token-level labeling tasks such as NER, it would be time-consuming to enumerate the template queries over all potential entity spans. In this work, we propose a more elegant method to reformulate NER tasks as LM problems without any templates. Specifically, we discard the template construction process while maintaining the word prediction paradigm of pre-training models to predict a class-related pivot word (or label word) at the entity position. Meanwhile, we also explore principled ways to automatically search for appropriate label words that the pre-trained models can easily adapt to. While avoiding complicated template-based process, the proposed LM objective also reduces the gap between different objectives used in pre-training and fine-tuning, thus it can better benefit the few-shot performance. Experimental results demonstrate the effectiveness of the proposed method over bert-tagger and template-based method under few-shot setting. Moreover, the decoding speed of the proposed method is up to 1930.12 times faster than the template-based method.