A membership inference attack (MIA) poses privacy risks for the training data of a machine learning model. With an MIA, an attacker guesses if the target data are a member of the training dataset. The state-of-the-art defense against MIAs, distillation for membership privacy (DMP), requires not only private data for protection but a large amount of unlabeled public data. However, in certain privacy-sensitive domains, such as medicine and finance, the availability of public data is not guaranteed. Moreover, a trivial method for generating public data by using generative adversarial networks significantly decreases the model accuracy, as reported by the authors of DMP. To overcome this problem, we propose a novel defense against MIAs that uses knowledge distillation without requiring public data. Our experiments show that the privacy protection and accuracy of our defense are comparable to those of DMP for the benchmark tabular datasets used in MIA research, Purchase100 and Texas100, and our defense has a much better privacy-utility trade-off than those of the existing defenses that also do not use public data for the image dataset CIFAR10.

We study algorithms for learning low-rank neural networks -- networks where the weight parameters are re-parameterized by products of two low-rank matrices. First, we present a provably efficient algorithm which learns an optimal low-rank approximation to a single-hidden-layer ReLU network up to additive error $\epsilon$ with probability $\ge 1 - \delta$, given access to noiseless samples with Gaussian marginals in polynomial time and samples. Thus, we provide the first example of an algorithm which can efficiently learn a neural network up to additive error without assuming the ground truth is realizable. To solve this problem, we introduce an efficient SVD-based $\textit{Nonlinear Kernel Projection}$ algorithm for solving a nonlinear low-rank approximation problem over Gaussian space. Inspired by the efficiency of our algorithm, we propose a novel low-rank initialization framework for training low-rank $\textit{deep}$ networks, and prove that for ReLU networks, the gap between our method and existing schemes widens as the desired rank of the approximating weights decreases, or as the dimension of the inputs increases (the latter point holds when network width is superlinear in dimension). Finally, we validate our theory by training ResNet and EfficientNet models on ImageNet.

Molecular conformation generation aims to generate three-dimensional coordinates of all the atoms in a molecule and is an important task in bioinformatics and pharmacology. Previous distance-based methods first predict interatomic distances and then generate conformations based on them, which could result in conflicting distances. In this work, we propose a method that directly predicts the coordinates of atoms. We design a dedicated loss function for conformation generation, which is invariant to roto-translation of coordinates of conformations and permutation of symmetric atoms in molecules. We further design a backbone model that stacks multiple blocks, where each block refines the conformation generated by its preceding block. Our method achieves state-of-the-art results on four public benchmarks: on small-scale GEOM-QM9 and GEOM-Drugs which have $200$K training data, we can improve the previous best matching score by $3.5\%$ and $28.9\%$; on large-scale GEOM-QM9 and GEOM-Drugs which have millions of training data, those two improvements are $47.1\%$ and $36.3\%$. This shows the effectiveness of our method and the great potential of the direct approach. Our code is released at \url{https://github.com/DirectMolecularConfGen/DMCG}.

Surrogate gradient (SG) training provides the possibility to quickly transfer all the gains made in deep learning to neuromorphic computing and neuromorphic processors, with the consequent reduction in energy consumption. Evidence supports that training can be robust to the choice of SG shape, after an extensive search of hyper-parameters. However, random or grid search of hyper-parameters becomes exponentially unfeasible as we consider more hyper-parameters. Moreover, every point in the search can itself be highly time and energy consuming for large networks and large datasets. In this article we show how complex tasks and networks are more sensitive to SG choice. Secondly, we show how low dampening, high sharpness and low tail fatness are preferred. Thirdly, we observe that Glorot Uniform initialization is generally preferred by most SG choices, with variability in the results. We finally provide a theoretical solution to reduce the need of extensive gridsearch, to find SG shape and initializations that result in improved accuracy.

Although Transformer-based methods have significantly improved state-of-the-art results for long-term series forecasting, they are not only computationally expensive but more importantly, are unable to capture the global view of time series (e.g. overall trend). To address these problems, we propose to combine Transformer with the seasonal-trend decomposition method, in which the decomposition method captures the global profile of time series while Transformers capture more detailed structures. To further enhance the performance of Transformer for long-term prediction, we exploit the fact that most time series tend to have a sparse representation in well-known basis such as Fourier transform, and develop a frequency enhanced Transformer. Besides being more effective, the proposed method, termed as Frequency Enhanced Decomposed Transformer ({\bf FEDformer}), is more efficient than standard Transformer with a linear complexity to the sequence length. Our empirical studies with six benchmark datasets show that compared with state-of-the-art methods, FEDformer can reduce prediction error by $14.8\%$ and $22.6\%$ for multivariate and univariate time series, respectively. the code will be released soon.

Recent approaches of computer vision utilize deep learning methods as they perform quite well if training and testing domains follow the same underlying data distribution. However, it has been shown that minor variations in the images that occur when using these methods in the real world can lead to unpredictable errors. Transfer learning is the area of machine learning that tries to prevent these errors. Especially, approaches that augment image data using auxiliary knowledge encoded in language embeddings or knowledge graphs (KGs) have achieved promising results in recent years. This survey focuses on visual transfer learning approaches using KGs. KGs can represent auxiliary knowledge either in an underlying graph-structured schema or in a vector-based knowledge graph embedding. Intending to enable the reader to solve visual transfer learning problems with the help of specific KG-DL configurations we start with a description of relevant modeling structures of a KG of various expressions, such as directed labeled graphs, hypergraphs, and hyper-relational graphs. We explain the notion of feature extractor, while specifically referring to visual and semantic features. We provide a broad overview of knowledge graph embedding methods and describe several joint training objectives suitable to combine them with high dimensional visual embeddings. The main section introduces four different categories on how a KG can be combined with a DL pipeline: 1) Knowledge Graph as a Reviewer; 2) Knowledge Graph as a Trainee; 3) Knowledge Graph as a Trainer; and 4) Knowledge Graph as a Peer. To help researchers find evaluation benchmarks, we provide an overview of generic KGs and a set of image processing datasets and benchmarks including various types of auxiliary knowledge. Last, we summarize related surveys and give an outlook about challenges and open issues for future research.

When explaining AI behavior to humans, how is the communicated information being comprehended by the human explainee, and does it match what the explanation attempted to communicate? When can we say that an explanation is explaining something? We aim to provide an answer by leveraging theory of mind literature about the folk concepts that humans use to understand behavior. We establish a framework of social attribution by the human explainee, which describes the function of explanations: the concrete information that humans comprehend from them. Specifically, effective explanations should be coherent (communicate information which generalizes to other contrast cases), complete (communicating an explicit contrast case, objective causes, and subjective causes), and interactive (surfacing and resolving contradictions to the generalization property through iterations). We demonstrate that many XAI mechanisms can be mapped to folk concepts of behavior. This allows us to uncover their modes of failure that prevent current methods from explaining effectively, and what is necessary to enable coherent explanations.

With rising uncertainty in the real world, online Reinforcement Learning (RL) has been receiving increasing attention due to its fast learning capability and improving data efficiency. However, online RL often suffers from complex Value Function Approximation (VFA) and catastrophic interference, creating difficulty for the deep neural network to be applied to an online RL algorithm in a fully online setting. Therefore, a simpler and more adaptive approach is introduced to evaluate value function with the kernel-based model. Sparse representations are superior at handling interference, indicating that competitive sparse representations should be learnable, non-prior, non-truncated and explicit when compared with current sparse representation methods. Moreover, in learning sparse representations, attention mechanisms are utilized to represent the degree of sparsification, and a smooth attentive function is introduced into the kernel-based VFA. In this paper, we propose an Online Attentive Kernel-Based Temporal Difference (OAKTD) algorithm using two-timescale optimization and provide convergence analysis of our proposed algorithm. Experimental evaluations showed that OAKTD outperformed several Online Kernel-based Temporal Difference (OKTD) learning algorithms in addition to the Temporal Difference (TD) learning algorithm with Tile Coding on public Mountain Car, Acrobot, CartPole and Puddle World tasks.

Detecting whether examples belong to a given in-distribution or are Out-Of-Distribution (OOD) requires identifying features specific to the in-distribution. In the absence of labels, these features can be learned by self-supervised techniques under the generic assumption that the most abstract features are those which are statistically most over-represented in comparison to other distributions from the same domain. In this work, we show that self-distillation of the in-distribution training set together with contrasting against negative examples derived from shifting transformation of auxiliary data strongly improves OOD detection. We find that this improvement depends on how the negative samples are generated. In particular, we observe that by leveraging negative samples, which keep the statistics of low-level features while changing the high-level semantics, higher average detection performance is obtained. Furthermore, good negative sampling strategies can be identified from the sensitivity of the OOD detection score. The efficiency of our approach is demonstrated across a diverse range of OOD detection problems, setting new benchmarks for unsupervised OOD detection in the visual domain.

Human infants learn language while interacting with their environment in which their caregivers may describe the objects and actions they perform. Similar to human infants, artificial agents can learn language while interacting with their environment. In this work, first, we present a neural model that bidirectionally binds robot actions and their language descriptions in a simple object manipulation scenario. Building on our previous Paired Variational Autoencoders (PVAE) model, we demonstrate the superiority of the variational autoencoder over standard autoencoders by experimenting with cubes of different colours, and by enabling the production of alternative vocabularies. Additional experiments show that the model's channel-separated visual feature extraction module can cope with objects of different shapes. Next, we introduce PVAE-BERT, which equips the model with a pretrained large-scale language model, i.e., Bidirectional Encoder Representations from Transformers (BERT), enabling the model to go beyond comprehending only the predefined descriptions that the network has been trained on; the recognition of action descriptions generalises to unconstrained natural language as the model becomes capable of understanding unlimited variations of the same descriptions. Our experiments suggest that using a pretrained language model as the language encoder allows our approach to scale up for real-world scenarios with instructions from human users.