Deep neural networks are usually trained in the space of the nodes, by adjusting the weights of existing links via suitable optimization protocols. We here propose a radically new approach which anchors the learning process to reciprocal space. Specifically, the training acts on the spectral domain and seeks to modify the eigenvalues and eigenvectors of transfer operators in direct space. The proposed method is ductile and can be tailored to return either linear or non-linear classifiers. Adjusting the eigenvalues, when freezing the eigenvectors entries, yields performances which are superior to those attained with standard methods {\it restricted} to a operate with an identical number of free parameters. Tuning the eigenvalues correspond in fact to performing a global training of the neural network, a procedure which promotes (resp. inhibits) collective modes on which an effective information processing relies. This is at variance with the usual approach to learning which implements instead a local modulation of the weights associated to pairwise links. Interestingly, spectral learning limited to the eigenvalues returns a distribution of the predicted weights which is close to that obtained when training the neural network in direct space, with no restrictions on the parameters to be tuned. Based on the above, it is surmised that spectral learning bound to the eigenvalues could be also employed for pre-training of deep neural networks, in conjunction with conventional machine-learning schemes. Changing the eigenvectors to a different non-orthogonal basis alters the topology of the network in direct space and thus allows to export the spectral learning strategy to other frameworks, as e.g. reservoir computing.

We investigate whisper-to-natural-speech conversion using sequence-to-sequence approach by proposing modified transformer architecture. We investigate different features like mel frequency cepstral coefficients and smoothed spectral features. The proposed networks are trained end-to-end using supervised approach for feature-to-feature transformation. Further, We also investigate the effectiveness of embedded auxiliary decoder used after N encoder sub-layers, and is trained with the frame-level objective function for identifying source phoneme labels. We show results on wTIMIT and CHAINS datasets by measuring word error rate using end-to-end ASR and also BLEU scores for the generated speech. In addition, we measure spectral shape of it by measuring formant distributions w.r.t. the reference speech, as formant divergence metric. We have found whisper-to-natural converted speech formants probability distribution is similar to the ground-truth distribution. To the authors' best knowledge, this is the first time modified transformer has been applied for whisper-to-natural-speech conversion and vice versa.

The migration of conversational AI agents across different embodiments in order to maintain the continuity of the task has been recently explored to further improve user experience. However, these migratable agents lack contextual understanding of the user information and the migrated device during the dialog conversations with the user. This opens the question of how an agent might behave when migrated into an embodiment for contextually predicting the next utterance. We collected a dataset from the dialog conversations between crowdsourced workers with the migration context involving personal and non-personal utterances in different settings (public or private) of embodiment into which the agent migrated. We trained the generative and information retrieval models on the dataset using with and without migration context and report the results of both qualitative metrics and human evaluation. We believe that the migration dataset would be useful for training future migratable AI systems.

Methods for understanding the decisions of and mechanisms underlying deep neural networks (DNNs) typically rely on building intuition by emphasizing sensory or semantic features of individual examples. For instance, methods aim to visualize the components of an input which are "important" to a network's decision, or to measure the semantic properties of single neurons. Here, we argue that interpretability research suffers from an over-reliance on intuition-based approaches that risk--and in some cases have caused--illusory progress and misleading conclusions. We identify a set of limitations that we argue impede meaningful progress in interpretability research, and examine two popular classes of interpretability methods--saliency and single-neuron-based approaches--that serve as case studies for how overreliance on intuition and lack of falsifiability can undermine interpretability research. To address these concerns, we propose a strategy to address these impediments in the form of a framework for strongly falsifiable interpretability research. We encourage researchers to use their intuitions as a starting point to develop and test clear, falsifiable hypotheses, and hope that our framework yields robust, evidence-based interpretability methods that generate meaningful advances in our understanding of DNNs.

This paper shows how to construct knowledge graphs (KGs) from pre-trained language models (e.g., BERT, GPT-2/3), without human supervision. Popular KGs (e.g, Wikidata, NELL) are built in either a supervised or semi-supervised manner, requiring humans to create knowledge. Recent deep language models automatically acquire knowledge from large-scale corpora via pre-training. The stored knowledge has enabled the language models to improve downstream NLP tasks, e.g., answering questions, and writing code and articles. In this paper, we propose an unsupervised method to cast the knowledge contained within language models into KGs. We show that KGs are constructed with a single forward pass of the pre-trained language models (without fine-tuning) over the corpora. We demonstrate the quality of the constructed KGs by comparing to two KGs (Wikidata, TAC KBP) created by humans. Our KGs also provide open factual knowledge that is new in the existing KGs. Our code and KGs will be made publicly available.

While the Transformer architecture has become the de-facto standard for natural language processing tasks, its applications to computer vision remain limited. In vision, attention is either applied in conjunction with convolutional networks, or used to replace certain components of convolutional networks while keeping their overall structure in place. We show that this reliance on CNNs is not necessary and a pure transformer applied directly to sequences of image patches can perform very well on image classification tasks. When pre-trained on large amounts of data and transferred to multiple mid-sized or small image recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains excellent results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train. Self-attention-based architectures, in particular Transformers (Vaswani et al., 2017), have become the model of choice in natural language processing (NLP). The dominant approach is to pre-train on a large text corpus and then fine-tune on a smaller task-specific dataset (Devlin et al., 2019). Thanks to Transformers' computational efficiency and scalability, it has become possible to train models of unprecedented size, with over 100B parameters. With the models and datasets growing, there is still no sign of saturating performance. In computer vision, however, convolutional architectures remain dominant (LeCun et al., 1989; Krizhevsky et al., 2012; He et al., 2016). Inspired by NLP successes, multiple works try combining CNN-like architectures with self-attention (Wang et al., 2018; Carion et al., 2020), some replacing the convolutions entirely (Ramachandran et al., 2019; Wang et al., 2020a). The latter models, while theoretically efficient, have not yet been scaled effectively on modern hardware accelerators due to the use of specialized attention patterns. Therefore, in large-scale image recognition, classic ResNetlike architectures are still state of the art (Mahajan et al., 2018; Xie et al., 2020; Kolesnikov et al., 2020). Inspired by the Transformer scaling successes in NLP, we experiment with applying a standard Transformer directly to images, with the fewest possible modifications. To do so, we split an image into patches and provide the sequence of linear embeddings of these patches as an input to a Transformer.

Most adversarial attack methods that are designed to deceive a text classifier change the text classifier's prediction by modifying a few words or characters. Few try to attack classifiers by rewriting a whole sentence, due to the difficulties inherent in sentence-level rephrasing as well as the problem of setting the criteria for legitimate rewriting. In this paper, we explore the problem of creating adversarial examples with sentence-level rewriting. We design a new sampling method, named ParaphraseSampler, to efficiently rewrite the original sentence in multiple ways. Then we propose a new criteria for modification, called a sentence-level threaten model. This criteria allows for both word- and sentence-level changes, and can be adjusted independently in two dimensions: semantic similarity and grammatical quality. Experimental results show that many of these rewritten sentences are misclassified by the classifier. On all 6 datasets, our ParaphraseSampler achieves a better attack success rate than our baseline.

It's been said that "Language Models are Unsupervised Multitask Learners." Indeed, self-supervised language models trained on "positive" examples of English text generalize in desirable ways to many natural language tasks. But if such models can stray so far from an initial self-supervision objective, a wayward model might generalize in undesirable ways too, say to nonsensical "negative" examples of unnatural language. A key question in this work is: do language models trained on (positive) training data also generalize to (negative) test data? We use this question as a contrivance to assess the extent to which language models learn undesirable properties of text, such as n-grams, that might interfere with the learning of more desirable properties of text, such as syntax. We find that within a model family, as the number of parameters, training epochs, and data set size increase, so does a model's ability to generalize to negative n-gram data, indicating standard self-supervision generalizes too far. We propose a form of inductive bias that attenuates such undesirable signals with negative data distributions automatically learned from positive data. We apply the method to remove n-gram signals from LSTMs and find that doing so causes them to favor syntactic signals, as demonstrated by large error reductions (up to 46% on the hardest cases) on a syntactic subject-verb agreement task.

Billions of text analysis requests containing private emails, personal text messages, and sensitive online reviews, are processed by recurrent neural networks (RNNs) deployed on public clouds every day. Although prior secure networks combine homomorphic encryption (HE) and garbled circuit (GC) to preserve users' privacy, naively adopting the HE and GC hybrid technique to implement RNNs suffers from long inference latency due to slow activation functions. In this paper, we present a HE and GC hybrid gated recurrent unit (GRU) network, CryptoGRU, for low-latency secure inferences. CryptoGRU replaces computationally expensive GC-based $tanh$ with fast GC-based $ReLU$, and then quantizes $sigmoid$ and $ReLU$ with a smaller bit length to accelerate activations in a GRU. We evaluate CryptoGRU with multiple GRU models trained on 4 public datasets. Experimental results show CryptoGRU achieves top-notch accuracy and improves the secure inference latency by up to $138\times$ over one of state-of-the-art secure networks on the Penn Treebank dataset.

Potential Drug-Drug Interaction(DDI) occurring while treating complex or co-existing diseases with drug combinations may cause changes in drugs' pharmacological activity. Therefore, DDI prediction has been an important task in the medical healthy machine learning community. Graph-based learning methods have recently aroused widespread interest and are proved to be a priority for this task. However, these methods are often limited to exploiting the inter-view drug molecular structure and ignoring the drug's intra-view interaction relationship, vital to capturing the complex DDI patterns. This study presents a new method, multi-view graph contrastive representation learning for drug-drug interaction prediction, MIRACLE for brevity, to capture inter-view molecule structure and intra-view interactions between molecules simultaneously. MIRACLE treats a DDI network as a multi-view graph where each node in the interaction graph itself is a drug molecular graph instance. We use GCN to encode DDI relationships and a bond-aware attentive message propagating method to capture drug molecular structure information in the MIRACLE learning stage. Also, we propose a novel unsupervised contrastive learning component to balance and integrate the multi-view information. Comprehensive experiments on multiple real datasets show that MIRACLE outperforms the state-of-the-art DDI prediction models consistently.