We describe SemEval-2017 Task 3 on Community Question Answering. This year, we reran the four subtasks from SemEval-2016:(A) Question-Comment Similarity,(B) Question-Question Similarity,(C) Question-External Comment Similarity, and (D) Rerank the correct answers for a new question in Arabic, providing all the data from 2015 and 2016 for training, and fresh data for testing. Additionally, we added a new subtask E in order to enable experimentation with Multi-domain Question Duplicate Detection in a larger-scale scenario, using StackExchange subforums. A total of 23 teams participated in the task, and submitted a total of 85 runs (36 primary and 49 contrastive) for subtasks A-D. Unfortunately, no teams participated in subtask E. A variety of approaches and features were used by the participating systems to address the different subtasks. The best systems achieved an official score (MAP) of 88.43, 47.22, 15.46, and 61.16 in subtasks A, B, C, and D, respectively. These scores are better than the baselines, especially for subtasks A-C.

Measuring conditional independence is one of the important tasks in statistical inference and is fundamental in causal discovery, feature selection, dimensionality reduction, Bayesian network learning, and others. In this work, we explore the connection between conditional independence measures induced by distances on a metric space and reproducing kernels associated with a reproducing kernel Hilbert space (RKHS). For certain distance and kernel pairs, we show the distance-based conditional independence measures to be equivalent to that of kernel-based measures. On the other hand, we also show that some popular---in machine learning---kernel conditional independence measures based on the Hilbert-Schmidt norm of a certain cross-conditional covariance operator, do not have a simple distance representation, except in some limiting cases. This paper, therefore, shows the distance and kernel measures of conditional independence to be not quite equivalent unlike in the case of joint independence as shown by Sejdinovic et al. (2013).

Training deep networks on light computational devices is nowadays very challenging. Continual learning techniques, where complex models are incrementally trained on small batches of new data, can make the learning problem tractable even for CPU-only edge devices. However, a number of practical problems need to be solved: catastrophic forgetting before anything else. In this paper we introduce an original technique named ``Latent Replay'' where, instead of storing a portion of past data in the input space, we store activations volumes at some intermediate layer. This can significantly reduce the computation and storage required by native rehearsal. To keep the representation stable and the stored activations valid we propose to slow-down learning at all the layers below the latent replay one, leaving the layers above free to learn at full pace. In our experiments we show that Latent Replay, combined with existing continual learning techniques, achieves state-of-the-art accuracy on a difficult benchmark such as CORe50 NICv2 with nearly 400 small and highly non-i.i.d. batches. Finally, we demonstrate the feasibility of nearly real-time continual learning on the edge through the porting of the proposed technique on a smartphone device.

Early detection and treatment of depression is essential in promoting remission, preventing relapse, and reducing the emotional burden of the disease. Current diagnoses are primarily subjective, inconsistent across professionals, and expensive for individuals who may be in urgent need of help. This paper proposes a novel approach to automated depression detection in speech using convolutional neural network (CNN) and multipart interactive training. The model was tested using 2568 voice samples obtained from 77 non-depressed and 30 depressed individuals. In experiment conducted, data were applied to residual CNNs in the form of spectrograms, images auto-generated from audio samples. The experimental results obtained using different ResNet architectures gave a promising baseline accuracy reaching 77%.

An ant colony optimization approach for partitioning a set of objects is proposed. In order to minimize the intra-variance, or within sum-of-squares, of the partitioned classes, we construct ant-like solutions by a constructive approach that selects objects to be put in a class with a probability that depends on the distance between the object and the centroid of the class (visibility) and the pheromone trail; the latter depends on the class memberships that have been defined along the iterations. The procedure is improved with the application of K-means algorithm in some iterations of the ant colony method. We performed a simulation study in order to evaluate the method with a Monte Carlo experiment that controls some sensitive parameters of the clustering problem. After some tuning of the parameters, the method has also been applied to some benchmark real-data sets. Encouraging results were obtained in nearly all cases.

Graphs possess exotic features like variable size and absence of natural ordering of the nodes that make them difficult to analyze and compare. To circumvent this problem and learn on graphs, graph feature representation is required. A good graph representation must satisfy the preservation of structural information, with two particular key attributes: consistency under deformation and invariance under isomorphism. While state-of-the-art methods seek such properties with powerful graph neural-networks, we propose to leverage a simple graph feature: the graph Laplacian spectrum (GLS). We first remind and show that GLS satisfies the aforementioned key attributes, using a graph perturbation approach. In particular, we derive bounds for the distance between two GLS that are related to the \textit{divergence to isomorphism}, a standard computationally expensive graph divergence. We finally experiment GLS as graph representation through consistency tests and classification tasks, and show that it is a strong graph feature representation baseline.

Bayesian inference problems require sampling or approximating high-dimensional probability distributions. The focus of this paper is on the recently introduced Stein variational gradient descent methodology, a class of algorithms that rely on iterated steepest descent steps with respect to a reproducing kernel Hilbert space norm. This construction leads to interacting particle systems, the mean-field limit of which is a gradient flow on the space of probability distributions equipped with a certain geometrical structure. We leverage this viewpoint to shed some light on the convergence properties of the algorithm, in particular addressing the problem of choosing a suitable positive definite kernel function. Our analysis leads us to considering certain nondifferentiable kernels with adjusted tails. We demonstrate significant performs gains of these in various numerical experiments.

Image normalization is a critical step in medical imaging. This step is often done on a per-dataset basis, preventing current segmentation algorithms from the full potential of exploiting jointly normalized information across multiple datasets. To solve this problem, we propose an adversarial normalization approach for image segmentation which learns common normalizing functions across multiple datasets while retaining image realism. The adversarial training provides an optimal normalizer that improves both the segmentation accuracy and the discrimination of unrealistic normalizing functions. Our contribution therefore leverages common imaging information from multiple domains. The optimality of our common normalizer is evaluated by combining brain images from both infants and adults. Results on the challenging iSEG and MRBrainS datasets reveal the potential of our adversarial normalization approach for segmentation, with Dice improvements of up to 59.6% over the baseline.

While state-of-the-art NLP explainability (XAI) methods focus on supervised, per-instance end or diagnostic probing task evaluation[4, 2, 10], this is insufficient to interpret and quantify model knowledge transfer during (un-) supervised training. By instead expressing each neuron as an interpretable token-activation distribution collected over many instances, one can quantify and guide visual exploration of neuron-knowledge change between model training stages to analyze transfer beyond probing tasks and the per-instance level. This allows one to analyze: (RQ1) how neurons abstract knowledge during unsupervised pretraining; (RQ2) how pretrained neurons zero-shot transfer knowledge to new domain data; and (RQ3) how supervised tasks reorder pretrained neuron knowledge abstractions. Since the meaningfulness of XAI methods is hard to quantify [11, 4], we analyze three example learning setups (RQ1-3) to empirically verify that our method (TX-Ray): identifies transfer (ir-)relevant neurons for pruning (RQ3), and that its transfer metrics coincide with traditional measures like perplexity (RQ1). We also find, that TX-Ray guided pruning of supervision (ir-)relevant neuron-knowledge (RQ3) can identify `lottery ticket'-like [9, 40] neurons that drive model performance and robustness. Upon inspecting pruned neurons, we find that task-relevant neuron-knowledge (`tickets'), appear (over-)fit, while task-irrelevant neurons lower overfitting, i.e. TX-Ray identifies neurons that generalize, transfer or specialize model-knowledge [25]. Finally, through RQ1-3, we find that TX-Ray helps to explore and quantify dynamics of (continual) knowledge transfer and that it can shed light on neuron-knowledge specialization and generalization, to complement (costly) supervised probing task procurement and established `summary' statistics like perplexity, ROC or F scores.

We tackle the task of environmental event classification by drawing inspiration from the transformer neural network architecture used in machine translation. We modify this attention-based feedforward structure in such a way that allows the resulting model to use audio as well as video to compute sound event predictions. We perform extensive experiments with these adapted transformers on an audiovisual data set, obtained by appending relevant visual information to an existing large-scale weakly labeled audio collection. The employed multi-label data contains clip-level annotation indicating the presence or absence of 17 classes of environmental sounds, and does not include temporal information. We show that the proposed modified transformers strongly improve upon previously introduced models and in fact achieve state-of-the-art results. We also make a compelling case for devoting more attention to research in multimodal audiovisual classification by proving the usefulness of visual information for the task at hand,namely audio event recognition. In addition, we visualize internal attention patterns of the audiovisual transformers and in doing so demonstrate their potential for performing multimodal synchronization.