This article is a summary of the paper by the European Union Agency for Fundamental Rights (FRA) called Data quality and artificial intelligence – mitigating bias and error to protect fundamental rights. I then proceed to look at the recent move by Facebook in its data politics; statements made by Zuckerberg; and their recent hiring of previous Deputy Prime Minister Nick Clegg as head of global policy and communications. It is my wish that this makes EU policy more comprehensible and give you an overview of a few actions taken by Facebook in this regard.
Combining social science and data science is not a new approach, yet after several revelations (and sizeable fines) large technology companies are waking up to discover where they are situated. It seems research institutes particularly in Europe are happy to facilitate this shift. This article is (1) a broad definition of data science; (2) a rapid look at social data science; (3) a surface look at how new, in relative terms, the discipline of social data science is at this moment.
Healthcare’s AI revolution is changing the way we think about age-related diseases, even aging itself We are in the midst of an epidemic. Regardless of your family history, race, or geography, there is a disease that will befall each and every one of us. You can hide in the mountains of Siberia, but the disease will still reach you because it’s not contagious. It’s followed humanity throughout time, and will continue to do so into the foreseeable future despite our recent attempts to forestall it. That disease is called aging.
It must be said that AI is wonderful when properly implemented. It must also be said that AI is frightening when unregulated. AI trailblazers are working towards establishing AI ethics – to varying success. Some early attempts have failed, such as Google’s attempt at establishing an AI ethics board earlier this year, which was dissolved after just a week. Instead, I argue that the future of establishing ethical AI lies in collaboration. For instance, the European Commission is inviting Europeans to discuss ethical AI. In my opinion, inviting a broad range of individuals and entities to establish ethical guidelines is the best approach to handling AI ethics. Hopefully, such initiatives will start to appear in greater scale in the near future. In order to ensure AI becomes – and stays – ethical, we must achieve diversified ethical boards through broad, inclusive discussions. Because someday soon, AI will decide whether you’re a criminal. And all you can do is to hope the AI makes the right call.
Artificial Intelligence (AI) is already having a major impact on society. As a result, many organizations have launched a wide range of initiatives to establish ethical principles for the adoption of socially beneficial AI. Unfortunately, the sheer volume of proposed principles threatens to overwhelm and confuse. How might this problem of ‘principle proliferation’ be solved? In this paper, we report the results of a fine-grained analysis of several of the highest-profile sets of ethical principles for AI. We assess whether these principles converge upon a set of agreed-upon principles, or diverge, with significant disagreement over what constitutes ‘ethical AI.’ Our analysis finds a high degree of overlap among the sets of principles we analyze. We then identify an overarching framework consisting of five core principles for ethical AI. Four of them are core principles commonly used in bioethics: beneficence, non-maleficence, autonomy, and justice. On the basis of our comparative analysis, we argue that a new principle is needed in addition: explicability, understood as incorporating both the epistemological sense of intelligibility (as an answer to the question ‘how does it work?’) and in the ethical sense of accountability (as an answer to the question: ‘who is responsible for the way it works?’). In the ensuing discussion, we note the limitations and assess the implications of this ethical framework for future efforts to create laws, rules, technical standards, and best practices for ethical AI in a wide range of contexts.
Michael Jordan’s article on artificial intelligence (AI) eloquently articulates how far we are from understanding human-level intelligence, much less recreating it through AI, machine learning, and robotics. The very premise that intelligent machines doing our work will make our lives better may be flawed. Evidence from neuroscience, cognitive science, health sciences, and gerontology shows that human wellbeing and longevity, our health and wellness, fundamentally hinge on physical activity, social connectedness, and a sense of purpose. Therefore, we may need very different types of AI from those currently in development to truly improve human quality of life at the individual and societal levels.
Microsoft is investing $1 billion in OpenAI to support us building artificial general intelligence (AGI) with widely distributed economic benefits. We’re partnering to develop a hardware and software platform within Microsoft Azure which will scale to AGI. We’ll jointly develop new Azure AI supercomputing technologies, and Microsoft will become our exclusive cloud provider – so we’ll be working hard together to further extend Microsoft Azure’s capabilities in large-scale AI systems.
The paper demonstrates that the matching estimator is not generally consistent for the average treatment effect of the treated when the matching is done without replacement using propensity scores. To achieve consistency, practitioners must either assume that no unit exists with a propensity score greater than one-half or assume that there is no confounding among such units. Illustrations suggest that the result applies also to matching using other metrics as long as it is done without replacement.
A growing number of methods aim to assess the challenging question of treatment effect variation in observational studies. This special section of ‘Observational Studies’ reports the results of a workshop conducted at the 2018 Atlantic Causal Inference Conference designed to understand the similarities and differences across these methods. We invited eight groups of researchers to analyze a synthetic observational data set that was generated using a recent large-scale randomized trial in education. Overall, participants employed a diverse set of methods, ranging from matching and flexible outcome modeling to semiparametric estimation and ensemble approaches. While there was broad consensus on the topline estimate, there were also large differences in estimated treatment effect moderation. This highlights the fact that estimating varying treatment effects in observational studies is often more challenging than estimating the average treatment effect alone. We suggest several directions for future work arising from this workshop.
Despite significant progress in dissecting the genetic architecture of complex diseases by genome-wide association studies (GWAS), the signals identified by association analysis may not have specific pathological relevance to diseases so that a large fraction of disease causing genetic variants is still hidden. Association is used to measure dependence between two variables or two sets of variables. Genome-wide association studies test association between a disease and SNPs (or other genetic variants) across the genome. Association analysis may detect superficial patterns between disease and genetic variants. Association signals provide limited information on the causal mechanism of diseases. The use of association analysis as a major analytical platform for genetic studies of complex diseases is a key issue that hampers discovery of the mechanism of diseases, calling into question the ability of GWAS to identify loci underlying diseases. It is time to move beyond association analysis toward techniques enabling the discovery of the underlying causal genetic strctures of complex diseases. To achieve this, we propose a concept of a genome-wide causation studies (GWCS) as an alternative to GWAS and develop additive noise models (ANMs) for genetic causation analysis. Type I error rates and power of the ANMs to test for causation are presented. We conduct GWCS of schizophrenia. Both simulation and real data analysis show that the proportion of the overlapped association and causation signals is small. Thus, we hope that our analysis will stimulate discussion of GWAS and GWCS.
How long should an online article title be? There’s a blog here citing an old post from 2013 which shows a nice plot for average click-through rate (CTR) and title length.
Mutual information has been successfully adopted in filter feature-selection methods to assess both the relevancy of a subset of features in predicting the target variable and the redundancy with respect to other variables. However, existing algorithms are mostly heuristic and do not offer any guarantee on the proposed solution. In this paper, we provide novel theoretical results showing that conditional mutual information naturally arises when bounding the ideal regression/classification errors achieved by different subsets of features. Leveraging on these insights, we propose a novel stopping condition for backward and forward greedy methods which ensures that the ideal prediction error using the selected feature subset remains bounded by a user-specified threshold. We provide numerical simulations to support our theoretical claims and compare to common heuristic methods.
The assumption of positivity in causal inference (also known as common support and co-variate overlap) is necessary to obtain valid causal estimates. Therefore, confirming it holds in a given dataset is an important first step of any causal analysis. Most common methods to date are insufficient for discovering non-positivity, as they do not scale for modern high-dimensional covariate spaces, or they cannot pinpoint the subpopulation violating positivity. To overcome these issues, we suggest to harness decision trees for detecting violations. By dividing the covariate space into mutually exclusive regions, each with maximized homogeneity of treatment groups, decision trees can be used to automatically detect subspaces violating positivity. By augmenting the method with an additional random forest model, we can quantify the robustness of the violation within each subspace. This solution is scalable and provides an interpretable characterization of the subspaces in which violations occur. We provide a visualization of the stratification rules that define each subpopulation, combined with the severity of positivity violation within it. We also provide an interactive version of the visualization that allows a deeper dive into the properties of each subspace.
Contextual Bilateral Loss (CoBi) This paper shows that when applying machine learning to digital zoom for photography, it is beneficial to use real, RAW sensor data for training. Existing learning-based super-resolution methods do not use real sensor data, instead operating on RGB images. In practice, these approaches result in loss of detail and accuracy in their digitally zoomed output when zooming in on distant image regions. We also show that synthesizing sensor data by resampling high-resolution RGB images is an oversimplified approximation of real sensor data and noise, resulting in worse image quality. The key barrier to using real sensor data for training is that ground truth high-resolution imagery is missing. We show how to obtain the ground-truth data with optically zoomed images and contribute a dataset, SR-RAW, for real-world computational zoom. We use SR-RAW to train a deep network with a novel contextual bilateral loss (CoBi) that delivers critical robustness to mild misalignment in input-output image pairs. The trained network achieves state-of-the-art performance in 4X and 8X computational zoom. …
Uncertainty-Aware Feature Selection (UAFS) Missing data are a concern in many real world data sets and imputation methods are often needed to estimate the values of missing data, but data sets with excessive missingness and high dimensionality challenge most approaches to imputation. Here we show that appropriate feature selection can be an effective preprocessing step for imputation, allowing for more accurate imputation and subsequent model predictions. The key feature of this preprocessing is that it incorporates uncertainty: by accounting for uncertainty due to missingness when selecting features we can reduce the degree of missingness while also limiting the number of uninformative features being used to make predictive models. We introduce a method to perform uncertainty-aware feature selection (UAFS), provide a theoretical motivation, and test UAFS on both real and synthetic problems, demonstrating that across a variety of data sets and levels of missingness we can improve the accuracy of imputations. Improved imputation due to UAFS also results in improved prediction accuracy when performing supervised learning using these imputed data sets. Our UAFS method is general and can be fruitfully coupled with a variety of imputation methods. …
Deep Feature Fusion-Audio and Text Modal Fusion (DFF-ATMF) Sentiment analysis research has been rapidly developing in the last decade and has attracted widespread attention from academia and industry, most of which is based on text. However, the information in the real world usually comes as different modalities. In this paper, we consider the task of Multimodal Sentiment Analysis, using Audio and Text Modalities, proposed a novel fusion strategy including Multi-Feature Fusion and Multi-Modality Fusion to improve the accuracy of Audio-Text Sentiment Analysis. We call this the Deep Feature Fusion-Audio and Text Modal Fusion (DFF-ATMF) model, and the features learned from it are complementary to each other and robust. Experiments with the CMU-MOSI corpus and the recently released CMU-MOSEI corpus for Youtube video sentiment analysis show the very competitive results of our proposed model. Surprisingly, our method also achieved the state-of-the-art results in the IEMOCAP dataset, indicating that our proposed fusion strategy is also extremely generalization ability to Multimodal Emotion Recognition. …
Distance Metric Learned Collaborative Representation Classifier (DML-CRC) Any generic deep machine learning algorithm is essentially a function fitting exercise, where the network tunes its weights and parameters to learn discriminatory features by minimizing some cost function. Though the network tries to learn the optimal feature space, it seldom tries to learn an optimal distance metric in the cost function, and hence misses out on an additional layer of abstraction. We present a simple effective way of achieving this by learning a generic Mahalanabis distance in a collaborative loss function in an end-to-end fashion with any standard convolutional network as the feature learner. The proposed method DML-CRC gives state-of-the-art performance on benchmark fine-grained classification datasets CUB Birds, Oxford Flowers and Oxford-IIIT Pets using the VGG-19 deep network. The method is network agnostic and can be used for any similar classification tasks. …
“And that’s where the statistician needs to take it easy:
• Start with the results, so the audience has a clear view on the outcome
• Proceed to explain the analysis simply and with a minimum of statistical jargon
• Describe what an algorithm does, not the specifics of your killer algo
• Visualize the inputs (e.g.: a correlation matrix showing an ‘influence heat map’)
• Visualize the process (e.g.: a regression line on a chief predictor variable)
• Visualize the results (e.g.: a lift chart to show how much the analysis is improving results)
• Always, always tie each step back to the business challenge
• Always be open to questions and feedback.”Andrew Pease ( November 3, 2014 )
The rapid growth of data in velocity, volume, value, variety, and veracity has enabled exciting new opportunities and presented big challenges for businesses of all types. Recently, there has been considerable interest in developing systems for processing continuous data streams with the increasing need for real-time analytics for decision support in the business, healthcare, manufacturing, and security. The analytics of streaming data usually relies on the output of offline analytics on static or archived data. However, businesses and organizations like our industry partner Gnowit, strive to provide their customers with real time market information and continuously look for a unified analytics framework that can integrate both streaming and offline analytics in a seamless fashion to extract knowledge from large volumes of hybrid streaming data. We present our study on designing a multilevel streaming text data analytics framework by comparing leading edge scalable open-source, distributed, and in-memory technologies. We demonstrate the functionality of the framework for a use case of multilevel text analytics using deep learning for language understanding and sentiment analysis including data indexing and query processing. Our framework combines Spark streaming for real time text processing, the Long Short Term Memory (LSTM) deep learning model for higher level sentiment analysis, and other tools for SQL-based analytical processing to provide a scalable solution for multilevel streaming text analytics.
Recently, collaborative robots have begun to train humans to achieve complex tasks, and the mutual information exchange between them can lead to successful robot-human collaborations. In this paper we demonstrate the application and effectiveness of a new approach called \textit{mutual reinforcement learning} (MRL), where both humans and autonomous agents act as reinforcement learners in a skill transfer scenario over continuous communication and feedback. An autonomous agent initially acts as an instructor who can teach a novice human participant complex skills using the MRL strategy. While teaching skills in a physical (block-building) () or simulated (Tetris) environment (), the expert tries to identify appropriate reward channels preferred by each individual and adapts itself accordingly using an exploration-exploitation strategy. These reward channel preferences can identify important behaviors of the human participants, because they may well exercise the same behaviors in similar situations later. In this way, skill transfer takes place between an expert system and a novice human operator. We divided the subject population into three groups and observed the skill transfer phenomenon, analyzing it with Simpson’s psychometric model. 5-point Likert scales were also used to identify the cognitive models of the human participants. We obtained a shared cognitive model which not only improves human cognition but enhances the robot’s cognitive strategy to understand the mental model of its human partners while building a successful robot-human collaborative framework.
Principles of cognitive economy would require that concepts about objects, properties and relations should be introduced only if they simplify the conceptualisation of a domain. Unexpectedly, classic logic conditionals, specifying structures holding within elements of a formal conceptualisation, do not always satisfy this crucial principle. The paper argues that this requirement is captured by \emph{supervenience}, hereby further identified as a property necessary for compression. The resulting theory suggests an alternative explanation of the empirical experiences observable in Wason’s selection tasks, associating human performance with conditionals on the ability of dealing with compression, rather than with logic necessity.
Improving the accuracy and robustness of deep neural nets (DNNs) and adapting them to small training data are primary tasks in deep learning research. In this paper, we replace the output activation function of DNNs, typically the data-agnostic softmax function, with a graph Laplacian-based high dimensional interpolating function which, in the continuum limit, converges to the solution of a Laplace-Beltrami equation on a high dimensional manifold. Furthermore, we propose end-to-end training and testing algorithms for this new architecture. The proposed DNN with graph interpolating activation integrates the advantages of both deep learning and manifold learning. Compared to the conventional DNNs with the softmax function as output activation, the new framework demonstrates the following major advantages: First, it is better applicable to data-efficient learning in which we train high capacity DNNs without using a large number of training data. Second, it remarkably improves both natural accuracy on the clean images and robust accuracy on the adversarial images crafted by both white-box and black-box adversarial attacks. Third, it is a natural choice for semi-supervised learning. For reproducibility, the code is available at \url{https://…/DNN-DataDependentActivation}.
Interpretable Machine Learning (IML) has become increasingly important in many applications, such as autonomous cars and medical diagnosis, where explanations are preferred to help people better understand how machine learning systems work and further enhance their trust towards systems. Particularly in robotics, explanations from IML are significantly helpful in providing reasons for those adverse and inscrutable actions, which could impair the safety and profit of the public. However, due to the diversified scenarios and subjective nature of explanations, we rarely have the ground truth for benchmark evaluation in IML on the quality of generated explanations. Having a sense of explanation quality not only matters for quantifying system boundaries, but also helps to realize the true benefits to human users in real-world applications. To benchmark evaluation in IML, in this paper, we rigorously define the problem of evaluating explanations, and systematically review the existing efforts. Specifically, we summarize three general aspects of explanation (i.e., predictability, fidelity and persuasibility) with formal definitions, and respectively review the representative methodologies for each of them under different tasks. Further, a unified evaluation framework is designed according to the hierarchical needs from developers and end-users, which could be easily adopted for different scenarios in practice. In the end, open problems are discussed, and several limitations of current evaluation techniques are raised for future explorations.
Knowledge tracing is the task of modeling each student’s mastery of knowledge concepts (KCs) as (s)he engages with a sequence of learning activities. Each student’s knowledge is modeled by estimating the performance of the student on the learning activities. It is an important research area for providing a personalized learning platform to students. In recent years, methods based on Recurrent Neural Networks (RNN) such as Deep Knowledge Tracing (DKT) and Dynamic Key-Value Memory Network (DKVMN) outperformed all the traditional methods because of their ability to capture complex representation of human learning. However, these methods face the issue of not generalizing well while dealing with sparse data which is the case with real-world data as students interact with few KCs. In order to address this issue, we develop an approach that identifies the KCs from the student’s past activities that are \textit{relevant} to the given KC and predicts his/her mastery based on the relatively few KCs that it picked. Since predictions are made based on relatively few past activities, it handles the data sparsity problem better than the methods based on RNN. For identifying the relevance between the KCs, we propose a self-attention based approach, Self Attentive Knowledge Tracing (SAKT). Extensive experimentation on a variety of real-world dataset shows that our model outperforms the state-of-the-art models for knowledge tracing, improving AUC by 4.43% on average.
Recommender systems are crucial to alleviate the information overload problem in online worlds. Most of the modern recommender systems capture users’ preference towards items via their interactions based on collaborative filtering techniques. In addition to the user-item interactions, social networks can also provide useful information to understand users’ preference as suggested by the social theories such as homophily and influence. Recently, deep neural networks have been utilized for social recommendations, which facilitate both the user-item interactions and the social network information. However, most of these models cannot take full advantage of the social network information. They only use information from direct neighbors, but distant neighbors can also provide helpful information. Meanwhile, most of these models treat neighbors’ information equally without considering the specific recommendations. However, for a specific recommendation case, the information relevant to the specific item would be helpful. Besides, most of these models do not explicitly capture the neighbor’s opinions to items for social recommendations, while different opinions could affect the user differently. In this paper, to address the aforementioned challenges, we propose DSCF, a Deep Social Collaborative Filtering framework, which can exploit the social relations with various aspects for recommender systems. Comprehensive experiments on two-real world datasets show the effectiveness of the proposed framework.
Recently, neural networks trained as optimizers under the ‘learning to learn’ or meta-learning framework have been shown to be effective for a broad range of optimization tasks including derivative-free black-box function optimization. Recurrent neural networks (RNNs) trained to optimize a diverse set of synthetic non-convex differentiable functions via gradient descent have been effective at optimizing derivative-free black-box functions. In this work, we propose RNN-Opt: an approach for learning RNN-based optimizers for optimizing real-parameter single-objective continuous functions under limited budget constraints. Existing approaches utilize an observed improvement based meta-learning loss function for training such models. We propose training RNN-Opt by using synthetic non-convex functions with known (approximate) optimal values by directly using discounted regret as our meta-learning loss function. We hypothesize that a regret-based loss function mimics typical testing scenarios, and would therefore lead to better optimizers compared to optimizers trained only to propose queries that improve over previous queries. Further, RNN-Opt incorporates simple yet effective enhancements during training and inference procedures to deal with the following practical challenges: i) Unknown range of possible values for the black-box function to be optimized, and ii) Practical and domain-knowledge based constraints on the input parameters. We demonstrate the efficacy of RNN-Opt in comparison to existing methods on several synthetic as well as standard benchmark black-box functions along with an anonymized industrial constrained optimization problem.
Deep learning techniques have become the method of choice for researchers working on algorithmic aspects of recommender systems. With the strongly increased interest in machine learning in general, it has, as a result, become difficult to keep track of what represents the state-of-the-art at the moment, e.g., for top-n recommendation tasks. At the same time, several recent publications point out problems in today’s research practice in applied machine learning, e.g., in terms of the reproducibility of the results or the choice of the baselines when proposing new models. In this work, we report the results of a systematic analysis of algorithmic proposals for top-n recommendation tasks. Specifically, we considered 18 algorithms that were presented at top-level research conferences in the last years. Only 7 of them could be reproduced with reasonable effort. For these methods, it however turned out that 6 of them can often be outperformed with comparably simple heuristic methods, e.g., based on nearest-neighbor or graph-based techniques. The remaining one clearly outperformed the baselines but did not consistently outperform a well-tuned non-neural linear ranking method. Overall, our work sheds light on a number of potential problems in today’s machine learning scholarship and calls for improved scientific practices in this area. Source code of our experiments and full results are available at: https://…/RecSys2019_DeepLearning_Evaluation.
We propose a quantum data fitting algorithm for non-sparse matrices, which is based on the Quantum Singular Value Estimation (QSVE) subroutine and a novel efficient method for recovering the signs of eigenvalues. Our algorithm generalizes the quantum data fitting algorithm of Wiebe, Braun, and Lloyd for sparse and well-conditioned matrices by adding a regularization term to avoid the over-fitting problem, which is a very important problem in machine learning. As a result, the algorithm achieves a sparsity-independent runtime of for an dimensional Hermitian matrix , where denotes the condition number of and is the precision parameter. This amounts to a polynomial speedup on the dimension of matrices when compared with the classical data fitting algorithms, and a strictly less than quadratic dependence on .
Numerosity perception is foundational to mathematical learning, but its computational bases are strongly debated. Some investigators argue that humans are endowed with a specialized system supporting numerical representation; others argue that visual numerosity is estimated using continuous magnitudes, such as density or area, which usually co-vary with number. Here we reconcile these contrasting perspectives by testing deep networks on the same numerosity comparison task that was administered to humans, using a stimulus space that allows to measure the contribution of non-numerical features. Our model accurately simulated the psychophysics of numerosity perception and the associated developmental changes: discrimination was driven by numerosity information, but non-numerical features had a significant impact, especially early during development. Representational similarity analysis further highlighted that both numerosity and continuous magnitudes were spontaneously encoded even when no task had to be carried out, demonstrating that numerosity is a major, salient property of our visual environment.
In a steady-state evolution, tournament selection traditionally uses the fitness function to select the parents, and negative selection chooses an individual to be replaced with an offspring. This contribution focuses on analyzing the behavior, in terms of performance, of different heuristics when used instead of the fitness function in tournament selection. The heuristics analyzed are related to measuring the similarity of the individuals in the semantic space. In addition, the analysis includes random selection and traditional tournament selection. These selection functions were implemented on our Semantic Genetic Programming system, namely EvoDAG, which is inspired by the geometric genetic operators and tested on 30 classification problems with a variable number of samples, variables, and classes. The result indicated that the combination of accuracy and the random selection, in the negative tournament, produces the best combination, and the difference in performances between this combination and the tournament selection is statistically significant. Furthermore, we compare EvoDAG’s performance using the selection heuristics against 18 classifiers that included traditional approaches as well as auto-machine-learning techniques. The results indicate that our proposal is competitive with state-of-art classifiers. Finally, it is worth to mention that EvoDAG is available as open source software.
We introduce natural adversarial examples — real-world, unmodified, and naturally occurring examples that cause classifier accuracy to significantly degrade. We curate 7,500 natural adversarial examples and release them in an ImageNet classifier test set that we call ImageNet-A. This dataset serves as a new way to measure classifier robustness. Like l_p adversarial examples, ImageNet-A examples successfully transfer to unseen or black-box classifiers. For example, on ImageNet-A a DenseNet-121 obtains around 2% accuracy, an accuracy drop of approximately 90%. Recovering this accuracy is not simple because ImageNet-A examples exploit deep flaws in current classifiers including their over-reliance on color, texture, and background cues. We observe that popular training techniques for improving robustness have little effect, but we show that some architectural changes can enhance robustness to natural adversarial examples. Future research is required to enable robust generalization to this hard ImageNet test set.
The presumed data owners’ right to explanations brought about by the General Data Protection Regulation in Europe has shed light on the social challenges of explainable artificial intelligence (XAI). In this paper, we present a case study with Deep Learning (DL) experts from a research and development laboratory focused on the delivery of industrial-strength AI technologies. Our aim was to investigate the social meaning (i.e. meaning to others) that DL experts assign to what they do, given a richly contextualized and familiar domain of application. Using qualitative research techniques to collect and analyze empirical data, our study has shown that participating DL experts did not spontaneously engage into considerations about the social meaning of machine learning models that they build. Moreover, when explicitly stimulated to do so, these experts expressed expectations that, with real-world DL application, there will be available mediators to bridge the gap between technical meanings that drive DL work, and social meanings that AI technology users assign to it. We concluded that current research incentives and values guiding the participants’ scientific interests and conduct are at odds with those required to face some of the scientific challenges involved in advancing XAI, and thus responding to the alleged data owners’ right to explanations or similar societal demands emerging from current debates. As a concrete contribution to mitigate what seems to be a more general problem, we propose three preliminary XAI Mediation Challenges with the potential to bring together technical and social meanings of DL applications, as well as to foster much needed interdisciplinary collaboration among AI and the Social Sciences researchers.
Multi-armed bandits a simple but very powerful framework for algorithms that make decisions over time under uncertainty. An enormous body of work has accumulated over the years, covered in several books and surveys. This book provides a more introductory, textbook-like treatment of the subject. Each chapter tackles a particular line of work, providing a self-contained, teachable technical introduction and a review of the more advanced results. The chapters are as follows: Stochastic bandits; Lower bounds; Bayesian Bandits and Thompson Sampling; Lipschitz Bandits; Full Feedback and Adversarial Costs; Adversarial Bandits; Linear Costs and Semi-bandits; Contextual Bandits; Bandits and Zero-Sum Games; Bandits with Knapsacks; Incentivized Exploration and Connections to Mechanism Design. Status of the manuscript: essentially complete (modulo some polishing), except for last chapter, which the author plans to add over the next few months.Introduction to Multi-Armed Bandits
Many teams try to start an applied AI project by diving into algorithms and data before figuring out desired outputs and objectives. Unfortunately, that’s like raising a puppy in a New York City apartment for a few years, then being surprised that it can’t herd sheep for you.
Many companies have adapted to a ‘data-driven’ approach for operational decision-making. Data can improve decisions, but it requires the right processor to get the most from it. Many people assume that processor is human. The term ‘data-driven’ even implies that data is curated by – and summarized for – people to process. But to fully leverage the value contained in data, companies need to bring artificial intelligence (AI) into their workflows and, sometimes, get us humans out of the way. We need to evolve from data-driven to AI-driven workflows. Distinguishing between ‘data-driven’ and ‘AI-driven’ isn’t just semantics. Each term reflects different assets, the former focusing on data and the latter processing ability. Data holds the insights that can enable better decisions; processing is the way to extract those insights and take actions. Humans and AI are both processors, with very different abilities. To understand how best to leverage each its helpful to review our own biological evolution and how decision-making has evolved in industry. Just fifty to seventy five years ago human judgment was the central processor of business decision-making. Professionals relied on their highly-tuned intuitions, developed from years of experience (and a relatively tiny bit of data) in their domain, to, say, pick the right creative for an ad campaign, determine the right inventory levels to stock, or approve the right financial investments. Experience and gut instinct were most of what was available to discern good from bad, high from low, and risky vs. safe.
Training deep neural networks to achieve the best performance is a challenging task. In this post, I would be exploring the most common problems and their solutions. These problems include taking too long to train, vanishing and exploding gradients and initialization. All these problems are known as Optimization problems. Another category of issue that arises while training the network is Regularization Problem. I have discussed them in my previous post. If you haven’t already read it, you can read it by clicking the link below.
I would like to begin by asking the following question: ‘Can we trust the model predictions just because the model performance is convincingly high on the test data?’ Many people might answer this question as ‘Yes’. But this is not always true. High model performance should not be considered an indicator to trust the model predictions, as the signals being picked up by the model can be random and might not make business sense.
I remember when I was having my first overseas internship at CERN as a summer student, most people were still talking about the discovery of Higgs boson upon confirming that it met the ‘five sigma’ threshold (which means having p-value of 0.0000003). Back then I knew nothing about p-value, hypothesis testing or even statistical significance. And you’re right. I went to google the word – p-value, and what I found on Wikipedia made me even more confused… In statistical hypothesis testing, the p-value or probability value is, for a given statistical model, the probability that, when the null hypothesis is true, the statistical summary (such as the absolute value of the sample mean difference between two compared groups) would be greater than or equal to the actual observed results.
In this post, we will discuss how to implement different combinations of non-linear activation functions and weight initialization methods in python. Also, we will analyze how the choice of activation function and weight initialization method will have an effect on accuracy and the rate at which we reduce our loss in a deep neural network using a non-linearly separable toy data set. This is a follow-up post to my previous post on activation functions and weight initialization methods. Note: This article assumes that the reader has a basic understanding of Neural Network, weights, biases, and backpropagation. If you want to learn the basics of the feed-forward neural network, check out my previous article (Link at the end of this article).
Computers are great at working with structured data like spreadsheets and database tables. But us humans usually communicate in words, not in tables. That’s unfortunate for computers.
I remember the first time I saw a computer, it was a Power Macintosh 5260 (with Monkey Island on it). I was around 5 years old and I looked at it as if it belonged to another universe. It did, I was not allowed to get anywhere close to it within a 5 mile radius; it was my older brother’s! That did not stop me. I browsed it for hours. The possibilities of computers were infinite and fuelled by the inspiration of sci-fi worlds the dream of talking machines, machines that can assist humans, think themselves and even have feelings never stopped. I kept dreaming about the possibilities of the future.
1. Enter ‘Generation Kaggle’
2. Neural Networks are the cure to everything
3. Machine Learning is the Product
4. Confuse Causation with Correlation
5. Optimize the wrong metrics
The Unity Machine Learning Agents Toolkit (ML-Agents) is an open-source Unity plugin that enables games and simulations to serve as environments for training intelligent agents. Agents can be trained using reinforcement learning, imitation learning, neuroevolution, or other machine learning methods through a simple-to-use Python API. We also provide implementations (based on TensorFlow) of state-of-the-art algorithms to enable game developers and hobbyists to easily train intelligent agents for 2D, 3D and VR/AR games. These trained agents can be used for multiple purposes, including controlling NPC behavior (in a variety of settings such as multi-agent and adversarial), automated testing of game builds and evaluating different game design decisions pre-release. The ML-Agents toolkit is mutually beneficial for both game developers and AI researchers as it provides a central platform where advances in AI can be evaluated on Unity’s rich environments and then made accessible to the wider research and game developer communities.
Hello everyone. I’m so excited to be here at Compose, along so many enthusiastic and some very advanced functional programmers. I live a dual life. At day, I teach computers how to think about code more deeply and, at night, I teach people how to think about code more deeply. So, this is the talk I’ve been really excited about the for last year; this is hands down the coolest thing I learned in the year of 2018. I was just reading this paper about programming language semantics and it was like, ‘Oh, these two things look completely different! Here’s how they’re the same, you do this.’ I was like: wait, what was that? What? It explains so many changes that I see people like myself already do. This got all of one slide in my web course I teach, but now I’ll get a chance to really explain why it’s so cool You are all here to learn.
Are you fascinated by the amount of text data available on the internet? Are you looking for ways to work with this text data but aren’t sure where to begin? Machines, after all, recognize numbers, not the letters of our language. And that can be a tricky landscape to navigate in machine learning.
1. Tokenization using Python’s split() function
2. Tokenization using Regular Expressions (RegEx)
3. Tokenization using NLTK
4. Tokenization using the spaCy library
5. Tokenization using Keras
6. Tokenization using Gensim
1. Machine Learning Use Cases in Smartphones
• Voice Assistants
• Smartphone Cameras
• App Store and Play Store Recommendations
• Face Unlock – Smartphones
2. Machine Learning Use Cases in Transportation
• Dynamic Pricing in Travel
• Transportation and Commuting – Uber
• Google Maps
3. Machine Learning Use Cases in Popular Web Services
• Email filtering
• Google Search
• Google Translate
• LinkedIn and Facebook recommendations and ads
4. Machine Learning Use Cases in Sales and Marketing
• Recommendation Engines
• Personalized Marketing
• Customer Support Queries (and Chatbots)
6. Machine Learning Use Cases in Security
• Video Surveillance
• Cyber Security (Captchas)
7. Machine Learning Use Cases in the Financial Domain
• Catching Fraud in Banking
• Personalized Banking
8. Other Popular Machine Learning Use Cases
• Self-Driving Cars
Ed Lorenz was a genius at coming up with simple models that capture the essence of a problem in a much more complex system. His famous butterfly model from 1963 jump-started chaos research, followed by more sophisticated models to describe upscale error growth (1969) and the general circulation of the atmosphere (1984). In 1995, he created another chaotic mode that shall be the topic of this blog post. Confusingly, even though the original paper appeared in 1995, most people refer to the model as the Lorenz 96 (L96) model, which we will also do here.
We’re not far from the day when artificial intelligence will provide us with a paintbrush for reality. As the foundations we’ve relied upon lose their integrity, many people find themselves afraid of what’s to come. But we’ve always lived in a world where our senses misrepresent reality. New technologies will help us get closer to the truth by showing us where we can’t find it. From a historical viewpoint, we’ve never successfully stopped the progression of any technology and owe the level of safety and security we enjoy to that ongoing progression. While normal accidents do occur and the downsides of progress likely won’t ever cease to exist, we make the problem worse when trying to fight the inevitable. Besides, reality has never been as clear and accurate as we want to believe. We fight against new technology because we believe it creates uncertainty when, more accurately, it only shines a light on the uncertainty that’s always existed and we’ve preferred to ignore.
At the end of June, Motherboard reported on a new app called DeepNude, which promised – ‘with a single click’ – to transform a clothed photo of any woman into a convincing nude image using machine learning. In the weeks since this report, the app has been pulled by its creator and removed from GitHub, though open source copies have surfaced there in recent days. Most of the coverage of DeepNude has focused on the specific dangers posed by its technical advances. ‘DeepNude is an evolution of that technology that is easier to use and faster to create than deepfakes,’ wrote Samantha Cole in Motherboard’s initial report on the app. ‘DeepNude also dispenses with the idea that this technology can be used for anything other than claiming ownership over women’s bodies.’ With its promise of single-click undressing of any woman, it made it easier than ever to manufacture naked photos – and, by extension, to use those fake nudes to harass, extort, and publicly shame women everywhere. But even following the app’s removal, there’s a lingering problem with DeepNude that goes beyond its technical advances and ease of use. It’s something older and deeper, something far more intractable – and far harder to erase from the internet – than a piece of open source code.
We recount in this essay the decade-long story of Gram Vaani, a social enterprise with a vision to build appropriate ICTs (Information and Communication Technologies) for participatory media in rural and low-income settings, to bring about social development and community empowerment. Other social enterprises will relate to the learning gained and the strategic pivots that Gram Vaani had to undertake to survive and deliver on its mission, while searching for a robust financial sustainability model. While we believe the ideal model still remains elusive, we conclude this essay with an open question about the reason to differentiate between different kinds of enterprises – commercial or social, for-profit or not-for-profit – and argue that all enterprises should have an ethical underpinning to their work.
With a view towards understanding why undesirable outcomes often arise in ICT projects, we draw attention to three aspects in this essay. First, we present several examples to show that incorporating an ethical framework in the design of an ICT system is not sufficient in itself, and that ethics need to guide the deployment and ongoing management of the projects as well. We present a framework that brings together the objectives, design, and deployment management of ICT projects as being shaped by a common underlying ethical system. Second, we argue that power-based equality should be incorporated as a key underlying ethical value in ICT projects, to ensure that the project does not reinforce inequalities in power relationships between the actors directly or indirectly associated with the project. We present a method to model ICT projects to make legible its influence on the power relationships between various actors in the ecosystem. Third, we discuss that the ethical values underlying any ICT project ultimately need to be upheld by the project teams, where certain factors like political ideologies or dispersed teams may affect the rigour with which these ethical values are followed. These three aspects of having an ethical underpinning to the design and management of ICT projects, the need for having a power-based equality principle for ICT projects, and the importance of socialization of the project teams, needs increasing attention in today’s age of ICT platforms where millions and billions of users interact on the same platform but which are managed by only a few people.
The presumed data owners’ right to explanations brought about by the General Data Protection Regulation in Europe has shed light on the social challenges of explainable artificial intelligence (XAI). In this paper, we present a case study with Deep Learning (DL) experts from a research and development laboratory focused on the delivery of industrial-strength AI technologies. Our aim was to investigate the social meaning (i.e. meaning to others) that DL experts assign to what they do, given a richly contextualized and familiar domain of application. Using qualitative research techniques to collect and analyze empirical data, our study has shown that participating DL experts did not spontaneously engage into considerations about the social meaning of machine learning models that they build. Moreover, when explicitly stimulated to do so, these experts expressed expectations that, with real-world DL application, there will be available mediators to bridge the gap between technical meanings that drive DL work, and social meanings that AI technology users assign to it. We concluded that current research incentives and values guiding the participants’ scientific interests and conduct are at odds with those required to face some of the scientific challenges involved in advancing XAI, and thus responding to the alleged data owners’ right to explanations or similar societal demands emerging from current debates. As a concrete contribution to mitigate what seems to be a more general problem, we propose three preliminary XAI Mediation Challenges with the potential to bring together technical and social meanings of DL applications, as well as to foster much needed interdisciplinary collaboration among AI and the Social Sciences researchers.
Introduction: To improve current public health strategies in suicide prevention and mental health, governments, researchers and private companies increasingly use information and communication technologies, and more specifically Artificial Intelligence and Big Data. These technologies are promising but raise ethical challenges rarely covered by current legal systems. It is essential to better identify, and prevent potential ethical risks. Objectives: The Canada Protocol – MHSP is a tool to guide and support professionals, users, and researchers using AI in mental health and suicide prevention. Methods: A checklist was constructed based upon ten international reports on AI and ethics and two guides on mental health and new technologies. 329 recommendations were identified, of which 43 were considered as applicable to Mental Health and AI. The checklist was validated, using a two round Delphi Consultation. Results: 16 experts participated in the first round of the Delphi Consultation and 8 participated in the second round. Of the original 43 items, 38 were retained. They concern five categories: ‘Description of the Autonomous Intelligent System’ (n=8), ‘Privacy and Transparency’ (n=8), ‘Security’ (n=6), ‘Health-Related Risks’ (n=8), ‘Biases’ (n=8). The checklist was considered relevant by most users, and could need versions tailored to each category of target users.
Online participatory media platforms that enable one-to-many communication among users, see a significant amount of user generated content and consequently face a problem of being able to recommend a subset of this content to its users. We address the problem of recommending and ranking this content such that different viewpoints about a topic get exposure in a fair and diverse manner. We build our model in the context of a voice-based participatory media platform running in rural central India, for low-income and less-literate communities, that plays audio messages in a ranked list to users over a phone call and allows them to contribute their own messages. In this paper, we describe our model and evaluate it using call-logs from the platform, to compare the fairness and diversity performance of our model with the manual editorial processes currently being followed. Our models are generic and can be adapted and applied to other participatory media platforms as well.
The ethical implications and social impacts of artificial intelligence have become topics of compelling interest to industry, researchers in academia, and the public. However, current analyses of AI in a global context are biased toward perspectives held in the U.S., and limited by a lack of research, especially outside the U.S. and Western Europe. This article summarizes the key findings of a literature review of recent social science scholarship on the social impacts of AI and related technologies in five global regions. Our team of social science researchers reviewed more than 800 academic journal articles and monographs in over a dozen languages. Our review of the literature suggests that AI is likely to have markedly different social impacts depending on geographical setting. Likewise, perceptions and understandings of AI are likely to be profoundly shaped by local cultural and social context. Recent research in U.S. settings demonstrates that AI-driven technologies have a pattern of entrenching social divides and exacerbating social inequality, particularly among historically-marginalized groups. Our literature review indicates that this pattern exists on a global scale, and suggests that low- and middle-income countries may be more vulnerable to the negative social impacts of AI and less likely to benefit from the attendant gains. We call for rigorous ethnographic research to better understand the social impacts of AI around the world. Global, on-the-ground research is particularly critical to identify AI systems that may amplify social inequality in order to mitigate potential harms. Deeper understanding of the social impacts of AI in diverse social settings is a necessary precursor to the development, implementation, and monitoring of responsible and beneficial AI technologies, and forms the basis for meaningful regulation of these technologies.
Failure to account for human values in software (e.g., equality and fairness) can result in user dissatisfaction and negative socio-economic impact. Engineering these values in software, however, requires technical and methodological support throughout the development life cycle. This paper investigates to what extent software engineering (SE) research has considered human values. We investigate the prevalence of human values in recent (2015 – 2018) publications at some of the top-tier SE conferences and journals. We classify SE publications, based on their relevance to different values, against a widely used value structure adopted from social sciences. Our results show that: (a) only a small proportion of the publications directly consider values, classified as relevant publications; (b) for the majority of the values, very few or no relevant publications were found; and (c) the prevalence of the relevant publications was higher in SE conferences compared to SE journals. This paper shares these and other insights that motivate research on human values in software engineering.
) or simulated (Tetris) environment (
), the expert tries to identify appropriate reward channels preferred by each individual and adapts itself accordingly using an exploration-exploitation strategy. These reward channel preferences can identify important behaviors of the human participants, because they may well exercise the same behaviors in similar situations later. In this way, skill transfer takes place between an expert system and a novice human operator. We divided the subject population into three groups and observed the skill transfer phenomenon, analyzing it with Simpson’s psychometric model. 5-point Likert scales were also used to identify the cognitive models of the human participants. We obtained a shared cognitive model which not only improves human cognition but enhances the robot’s cognitive strategy to understand the mental model of its human partners while building a successful robot-human collaborative framework.
for an 
dimensional Hermitian matrix 
, where 
denotes the condition number of 
is the precision parameter. This amounts to a polynomial speedup on the dimension of matrices when compared with the classical data fitting algorithms, and a strictly less than quadratic dependence on