Speech Image Semantic Alignment Does Not Depend on Any Prior Classification Tasks

Semantically-aligned $(speech, image)$ datasets can be used to explore”visually-grounded speech” Previous results have tended to show low rates of recall in $speech\rightarrow image$ and $image \rightarrow speech$ queries . Choosing appropriate neural architectures for encoders in the speech andimage branches and using large datasets, one can obtain competitive recall rates without any reliance on any pre-trained initialization or featureextraction .…

Capacity achieving codes a review on double transitivity

If a linear code is invariant under the action of a doubly transitive permutation group, it achieves the capacity of erasurechannel . Therefore, it is of sufficient interest to classify all codes,invariant under such permutation groups . We take a step in this direction and give a review of all suitable groups and the known results on codes .…

Lessons Learned from the 1st ARIEL Machine Learning Challenge Correcting Transiting Exoplanet Light Curves for Stellar Spots

Machine Learning Challenge was organized for the EuropeanSpace Agency’s upcoming Ariel mission . Successful solutions either construct highlynon-linear (w.r.t. the raw data) models with minimal preprocessing -deep neuralnetworks and ensemble methods- or amount to obtaining meaningful statistics from the light curves, constructing linear models on which yields comparably good predictive performance.…

Greedy Optimization Provably Wins the Lottery Logarithmic Number of Winning Tickets is Enough

The proposed method has the guarantee that the discrepancy between the pruned network and the original network decays with exponentiallyfast rate w.r.t. the size of the prune network, under weak assumptions that apply for most practical settings . Empirically, our method improves prior artson pruning various network architectures including ResNet, MobilenetV2/V3 onImageNet .…

Entanglement Induced Barren Plateaus

We argue that an excess in entanglement between the visible and hidden units in a Quantum Neural Network can hinder learning . We show that for any bounded objectivefunction on the visible layers, the Lipshitz constants of the expectation value of that objective function will scale inversely with the dimension of the hidden-subsystem with high probability.…

AutoPrompt Eliciting Knowledge from Language Models with Automatically Generated Prompts

The remarkable success of pretrained language models has motivated the study of what kinds of knowledge these models learn during pretraining . Reformulating tasks as fill-in-the-blanks problems (e.g., cloze tests) is a natural approach to gauging such knowledge . Using AutoPrompt, we show that maskedlanguage models (MLMs) have an inherent capability to perform sentimentanalysis and natural language inference without additional parameters orfinetuning .…

Model Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System

The aim of the recommender system is to provide personalized suggestions to users, not to suggest popular items . This results in the terrible Matthew effect, making popular items be more frequently recommended and become even more popular . To eliminate popularity bias, it is essential to answer the question that what the ranking score would be if the model onlyuses item property .…

Bayes Adaptive Deep Model Based Policy Optimisation

We introduce a Bayesian (deep) model-based reinforcement learning method(RoMBRL) that can capture model uncertainty to achieve sample-efficient policyoptimisation . We show that RoMBRL outperforms existing approaches on many challenging control benchmark tasks in terms of sample complexity and task performance .…

Human versus Machine Attention in Deep Reinforcement Learning Tasks

Deep reinforcement learning (RL) algorithms are powerful tools for solving motor motor decision tasks . However, the trained models are often difficult tointerpret, because they are represented as end-to-end deep neural networks . In this paper, we shed light on the inner workings of such trained models by analyzing the pixels that they attend to during task execution .…

PAL Pretext based Active Learning

Activelearning refers to the development of algorithms to judiciously pick limitedsubsets of unlabeled samples that can be sent for labeling by an oracle . When obtaining labels is expensive, the requirement of a large labeled training data set for deep learning can be mitigated by active learning .…

Fundamental limitations to key distillation from Gaussian states with Gaussian operations

We establish fundamental upper bounds on the amount of secret key that can be extracted from continuous variable quantum Gaussian states by using only localGaussian operations, local classical processing, and public communication . For one-way communication, we prove that the key is bounded by the R\’enyi-$2$Gaussian entanglement of formation $E_{F,2}^{\mathrm{\scriptscriptstyle G}}$ The same is true if two-way public communication is allowed but Alice and Bob employ protocolsthat start with destructive local Gaussian measurements .…

Differential Privacy and Natural Language Processing to Generate Contextually Similar Decoy Messages in Honey Encryption Scheme

Honey Encryption is an approach to encrypt the messages using low min-entropykeys, such as weak passwords, OTPs, PINs, credit card numbers . But the currenttechniques used in producing the decoy plaintexts do not model human languageentirely . A gibberish, random assortment of words is not enough to fool anattacker; that will not be acceptable and convincing, whether or not the attacker knows some information of the genuine source .…

Iteratively reweighted greedy set cover

We empirically analyze a simple heuristic for large sparse set coverproblems . It uses a weighted greedy algorithm as a basic building block . By multiplying updates of the weights attached to the elements, the greedysolution is improved . The implementation of this algorithm is trivial and the algorithm is essentially free of parameters that would requiretuning .…

Recursive Random Contraction Revisited

In this note, we revisit the recursive random contraction algorithm of Kargerand Stein for finding a minimum cut in a graph . We show that the analysis becomes particularly clean in the analysis of graphs . We also consider other similar variants of the algorithm, and show that no such algorithm can achieve an asymptotically better probability of finding a fixed minimum cut .…

Quantum advantage for differential equation analysis

Quantum algorithms for both differential equation solving and for machinelearning potentially offer an exponential speedup over all known classical algorithms . The essential obstacle for quantumdifferential equation solving is that outputting useful information may requiredifficult post-processing, and for quantum machinelearning is that inputting the training set is a difficult task just by itself .…

Graph Regularized Autoencoder and its Application in Unsupervised Anomaly Detection

Dimensionality reduction is a crucial first step for many unsupervised learning tasks including anomaly detection . We propose to use a minimum spanning tree (MST),a graph-based algorithm, to approximate the local neighborhood structure andgenerate structure-preserving distances among data points . We find that using the MST regularizer improves the performance of anomaly detection substantially forboth generative adversarial networks .…

Speech Based Emotion Recognition using Neural Networks and Information Visualization

Emotions recognition is commonly employed for health assessment . However, thetypical metric for evaluation in therapy is based on patient-doctor appraisal . Machine learning algorithms can be a useful tool for the classification ofemotions . We propose a tool which enables users to take speech samples and identify a range of emotions from audio elements through a machine learning model .…

papaya2 2D Irreducible Minkowski Tensor computation

Papaya2 is a software to calculate 2D higher-order shape metrics with a libraryinterface, support for Irreducible Minkowski Tensors and interpolated marchingsquares . Extensions to Matlab, JavaScript and Python are provided as well . We are not aware of other open-source software which provides higher-rank shape characterization in2D .…

Over parametrized neural networks as under determined linear systems

We draw connections between simple neural networks and under-determinedlinear systems to comprehensively explore several interesting theoreticalquestions in the study of neural networks . We emphatically show that itis unsurprising such networks can achieve zero training loss . Our lower bounds grow more slowly with data set size than existing work that trains the hidden layer weights .…

ACCDOA Activity Coupled Cartesian Direction of Arrival Representation for Sound Event Localization and Detection

Conventional NN-based methods use twobranches for a sound event detection (SED) target and a direction-of-arrival (DOA) target . Using two networks dedicated to each task increases system complexity and network size . We propose an activity-coupled Cartesian DOA (ACCDOA)representation . The ACCDOA representation enables us to solve a SELD task with a single target and has two advantages: avoiding thenecessity of balancing the objectives and model size increase .…

Unveiling process insights from refactoring practices

Software comprehension and maintenance activities, such as refactoring, are said to be negatively impacted by software complexity . Most teams using a plugin forrefactoring (JDeodorant) reduced software complexity more effectively and withsimpler processes than those using only Eclipsenative features . We were able to find moderate correlations (43%) between software cyclomatic complexity and process cyclomatic complexity .…

CoroBase Coroutine Oriented Main Memory Database Engine

Data stalls are a major overhead in main-memory database engines due to the use of pointer-rich data structures . Lightweight coroutines ease the implementation of software prefetching to hide data stalls . CoroBase can perform 2x better for read-intensive workloads and remain competitive for those workloads that do not benefit from software pre-etching .…

The IQIYI System for Voice Conversion Challenge 2020

IQIYI voice conversion system (T24) for VoiceConversion 2020 . In the competition, each target speaker has 70 sentences . The evaluation results show that this system canachieve better voice conversion effects. In the case of using 16k rather than 24k sampling rate audio, the conversion result is relatively good innaturalness and similarity.…

Discovery and classification of Twitter bots

Botnets may be used to betterinfiltrate the social graph over time and to create an illusion of community behavior, amplifying their message and increasing persuasion . We analyzed a dense crawl of a subset of Twitter traffic amounting to nearly all interactions by Greek-speaking Twitter users for a period of 36 months .…