Since 1930, Ad Age is your daily-must read covering the rapid changes taking place in media, retail, food and beverage, sports marketing, travel and more.

Taylor Swift's new album—how Sephora, Sweetgreen and other brands are 
reacting to 'The Tortured Poets Department'

According to its CEO, Sephora&#x27;s broad-based growth is a testament to its global success. How can Sephora balance localized strategies with its global brand identity for sustained growth?

Will Sephora's 4 Ingredients for Success Continue To Work? - RetailWire

At Sephora, it&#x27;s not just adults searching for the perfect beauty product. Children, some as young as 10, are now becoming regulars, inspired by social media influencers and driven by

Sephora Kids: A viral epidemic taking over young girls | News | tigertv.tv

The Metropolitan Police has launched a murder probe in south London after the death of the 26-year-old died of stab wounds.Separate stabbing incidents » · Conservative leadership hopeful Boris Johnson has defended his infamous description of black people in Africa bearing “watermelon smiles” ...

'Sephora kids' trend sparks health concerns

Sepana

sepana

Authors:
(1) Sebastian Dziadzio, University of Tübingen (sebastian.dziadzio@uni-tuebingen.de);
(2) Çagatay Yıldız, University of Tübingen;
(3) Gido M. van de Ven, KU Leuven;
(4) Tomasz Trzcinski, IDEAS NCBR, Warsaw University of Technology, Tooploox;
(5) Tinne Tuytelaars, KU Leuven;
(6) Matthias Bethge, University of Tübingen. Table of Links Abstract and 1. Introduction 2. Two problems with the current approach to class-incremental continual learning 3. Methods and 3.1. Infinite dSprites 3.2. Disentangled learning 4. Related work 4.1. Continual learning and 4.2. Benchmarking continual learning 5. Experiments 5.1. Regularization methods and 5.2. Replay-based methods 5.3. Do we need equivariance? 5.4. One-shot generalization and 5.5. Open-set classification 5.6. Online vs. offline Conclusion, Acknowledgments and References Supplementary Material Abstract The ability of machine learning systems to learn continually is hindered by catastrophic forgetting, the tendency of neural networks to overwrite existing knowledge when learning a new task. Existing continual learning methods alleviate this problem through regularisation, parameter isolation, or rehearsal, and are typically evaluated on benchmarks consisting of a handful of tasks. We propose a novel conceptual approach to continual classification that aims to disentangle class-specific information that needs to be memorized from the class-agnostic knowledge that encapsulates generalization. We store the former in a buffer that can be easily pruned or updated when new categories arrive, while the latter is represented with a neural network that generalizes across tasks. We show that the class-agnostic network does not suffer from catastrophic forgetting and by leveraging it to perform classification, we improve accuracy on past tasks over time. In addition, our approach supports open-set classification and one-shot generalization. To test our conceptual framework, we introduce Infinite dSprites, a tool for creating continual classification and disentanglement benchmarks of arbitrary length with full control over generative factors. We show that over a sufficiently long time horizon all major types of continual learning methods break down, while our approach enables continual learning over hundreds of tasks with explicit control over memorization and forgetting. 1. Introduction A machine learning system designed for continual learning must not only adapt to the current task, but also improve its performance on past tasks and build representations that facilitate the learning of future tasks. The latter two requirements are known as backward and forward transfer. The path to meeting these requirements is obstructed by catastrophic forgetting, the inability to preserve existing knowledge upon learning new information. As noted in early studies [24], catastrophic forgetting is caused by destructive model updates, where adjustments to model parameters, made through gradient descent, focus solely on the current task’s objective and can potentially impair performance on past tasks. To mitigate this issue, continual learning methods employ strategies such as (i) regularization, which aims to preserve existing knowledge by limiting the plasticity of selected network weights [15, 17, 26, 36], (ii) parameter isolation or dynamic architectures, which effectively solve each task with a dedicated model [6, 33], or (iii) replay, which augments the training data with stored samples from past tasks [4, 12, 30, 32]. Most continual learning methods are evaluated on image classification benchmarks in which a discriminative model is transferred across tasks that typically involve disjoint sets of classes. We argue that this purely discriminative learning framework is not conducive to positive forward or backward transfer. Supervised classification networks tend to preserve only the features that are relevant for predicting the output labels in the training data [11, 35]. In a continual learning setting, these features transfer poorly to future tasks with a completely different set of labels. Conversely, gradient updates with respect to current task’s objective do not encourage preserving features relevant to previous tasks. Based on these observations, we propose an alternative paradigm for continual learning centered around the idea of transferring modules that learn the general aspects of the problem (for example identity-preserving transforms that act similarly on all objects, such as illumination changes). We hypothesize that destructive model updates can be avoided by separating two objectives: (i) generalization, or learning class-agnostic transforms that successfully transfer to past and future tasks, and (ii) memorization of class-specific information. Crucially, our framework resolves the catastrophic forgetting issue by disentangled learning, that is, having a separate update procedure for the generalization model and the memory buffer (please note the difference from learning disentangled representations). This separation allows us to maintain, prune, and expand task-specific knowledge stored in the memory buffer while continuously training the generalization model. By focusing on learning the universal transformations, we can not only avoid destructive gradient updates, but efficiently accumulate knowledge over time. To demonstrate our proposed idea of learning universal transformations, we introduce infinite dSprites (idSprites), a continual learning benchmark generator inspired by the dSprites dataset [23]. It allows for procedurally generating a virtually infinite progression of random two-dimensional shapes. Similar to [23], we generate each unique shape in every combination of orientation, scale, and position (see Fig. 2). Most importantly, by providing the ground truth values of individual factors of variation (FoVs), idSprites enables us to learn general transformations, thereby separating generalization from memorization and testing our main hypothesis. We hope that by releasing idSprites as a Python package we will encourage the research community to test their methods on our benchmark. Section 3.2 describes an implementation of our disentangled learning framework in the context of class-incremental continual learning. Our proof of concept consists of an equivariant network that learns to regress the parameters of an affine transform that maps any input into its canonical form, a normalisation module that applies the predicted affine transformation to the input, and an exemplar buffer that stores a single exemplar per class. Figure 1 shows the main components of our framework. Contributions We summarize the most important contributions of this work below: • We introduce a new framework for generating continual classification and disentanglement benchmarks that for the first time allows testing continual learning methods over thousands of tasks. We will open-source our software package upon acceptance. • We propose a novel continual learning paradigm based on learning symmetry transformations, which circumvents catastrophic forgetting by separating gradient-based model updates from explicit memory edits. • We demonstrate that as the number of tasks grows, regularization-based continual learning methods quickly break down and replay-based methods either deteriorate in performance or become impractical due to extensive use of memory and compute. • We show that our approach exhibits significant forward and backward transfer, strong open-set classification performance, and excellent zero-shot generalisation. It can learn over hundreds of tasks with a constant computational budget and a slowly growing memory footprint. This paper is available on arxiv under CC 4.0 license. Authors: (1) Sebastian Dziadzio, University of Tübingen (sebastian.dziadzio@uni-tuebingen.de); (2) Çagatay Yıldız, University of Tübingen; (3) Gido M. van de Ven, KU Leuven; (4) Tomasz Trzcinski, IDEAS NCBR, Warsaw University of Technology, Tooploox; (5) Tinne Tuytelaars, KU Leuven; (6) Matthias Bethge, University of Tübingen. Authors: Authors: (1) Sebastian Dziadzio, University of Tübingen (sebastian.dziadzio@uni-tuebingen.de); (2) Çagatay Yıldız, University of Tübingen; (3) Gido M. van de Ven, KU Leuven; (4) Tomasz Trzcinski, IDEAS NCBR, Warsaw University of Technology, Tooploox; (5) Tinne Tuytelaars, KU Leuven; (6) Matthias Bethge, University of Tübingen. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Two problems with the current approach to class-incremental continual learning 2. Two problems with the current approach to class-incremental continual learning 3. Methods and 3.1. Infinite dSprites 3. Methods and 3.1. Infinite dSprites 3.2. Disentangled learning 3.2. Disentangled learning 4. Related work 4.1. Continual learning and 4.2. Benchmarking continual learning 4.1. Continual learning and 4.2. Benchmarking continual learning 5. Experiments 5. Experiments 5.1. Regularization methods and 5.2. Replay-based methods 5.1. Regularization methods and 5.2. Replay-based methods 5.3. Do we need equivariance? 5.3. Do we need equivariance? 5.4. One-shot generalization and 5.5. Open-set classification 5.4. One-shot generalization and 5.5. Open-set classification 5.6. Online vs. offline 5.6. Online vs. offline Conclusion, Acknowledgments and References Conclusion, Acknowledgments and References Supplementary Material Supplementary Material Abstract The ability of machine learning systems to learn continually is hindered by catastrophic forgetting, the tendency of neural networks to overwrite existing knowledge when learning a new task. Existing continual learning methods alleviate this problem through regularisation, parameter isolation, or rehearsal, and are typically evaluated on benchmarks consisting of a handful of tasks. We propose a novel conceptual approach to continual classification that aims to disentangle class-specific information that needs to be memorized from the class-agnostic knowledge that encapsulates generalization. We store the former in a buffer that can be easily pruned or updated when new categories arrive, while the latter is represented with a neural network that generalizes across tasks. We show that the class-agnostic network does not suffer from catastrophic forgetting and by leveraging it to perform classification, we improve accuracy on past tasks over time. In addition, our approach supports open-set classification and one-shot generalization. To test our conceptual framework, we introduce Infinite dSprites, a tool for creating continual classification and disentanglement benchmarks of arbitrary length with full control over generative factors. We show that over a sufficiently long time horizon all major types of continual learning methods break down, while our approach enables continual learning over hundreds of tasks with explicit control over memorization and forgetting. The ability of machine learning systems to learn continually is hindered by catastrophic forgetting, the tendency of neural networks to overwrite existing knowledge when learning a new task. Existing continual learning methods alleviate this problem through regularisation, parameter isolation, or rehearsal, and are typically evaluated on benchmarks consisting of a handful of tasks. We propose a novel conceptual approach to continual classification that aims to disentangle class-specific information that needs to be memorized from the class-agnostic knowledge that encapsulates generalization. We store the former in a buffer that can be easily pruned or updated when new categories arrive, while the latter is represented with a neural network that generalizes across tasks. We show that the class-agnostic network does not suffer from catastrophic forgetting and by leveraging it to perform classification, we improve accuracy on past tasks over time. In addition, our approach supports open-set classification and one-shot generalization. To test our conceptual framework, we introduce Infinite dSprites, a tool for creating continual classification and disentanglement benchmarks of arbitrary length with full control over generative factors. We show that over a sufficiently long time horizon all major types of continual learning methods break down, while our approach enables continual learning over hundreds of tasks with explicit control over memorization and forgetting. 1. Introduction A machine learning system designed for continual learning must not only adapt to the current task, but also improve its performance on past tasks and build representations that facilitate the learning of future tasks. The latter two requirements are known as backward and forward transfer. The path to meeting these requirements is obstructed by catastrophic forgetting, the inability to preserve existing knowledge upon learning new information. As noted in early studies [24], catastrophic forgetting is caused by destructive model updates, where adjustments to model parameters, made through gradient descent, focus solely on the current task’s objective and can potentially impair performance on past tasks. To mitigate this issue, continual learning methods employ strategies such as (i) regularization, which aims to preserve existing knowledge by limiting the plasticity of selected network weights [15, 17, 26, 36], (ii) parameter isolation or dynamic architectures, which effectively solve each task with a dedicated model [6, 33], or (iii) replay, which augments the training data with stored samples from past tasks [4, 12, 30, 32]. Most continual learning methods are evaluated on image classification benchmarks in which a discriminative model is transferred across tasks that typically involve disjoint sets of classes. We argue that this purely discriminative learning framework is not conducive to positive forward or backward transfer. Supervised classification networks tend to preserve only the features that are relevant for predicting the output labels in the training data [11, 35]. In a continual learning setting, these features transfer poorly to future tasks with a completely different set of labels. Conversely, gradient updates with respect to current task’s objective do not encourage preserving features relevant to previous tasks. Based on these observations, we propose an alternative paradigm for continual learning centered around the idea of transferring modules that learn the general aspects of the problem (for example identity-preserving transforms that act similarly on all objects, such as illumination changes). We hypothesize that destructive model updates can be avoided by separating two objectives: (i) generalization, or learning class-agnostic transforms that successfully transfer to past and future tasks, and (ii) memorization of class-specific information. Crucially, our framework resolves the catastrophic forgetting issue by disentangled learning, that is, having a separate update procedure for the generalization model and the memory buffer (please note the difference from learning disentangled representations). This separation allows us to maintain, prune, and expand task-specific knowledge stored in the memory buffer while continuously training the generalization model. By focusing on learning the universal transformations, we can not only avoid destructive gradient updates, but efficiently accumulate knowledge over time. To demonstrate our proposed idea of learning universal transformations, we introduce infinite dSprites (idSprites), a continual learning benchmark generator inspired by the dSprites dataset [23]. It allows for procedurally generating a virtually infinite progression of random two-dimensional shapes. Similar to [23], we generate each unique shape in every combination of orientation, scale, and position (see Fig. 2). Most importantly, by providing the ground truth values of individual factors of variation (FoVs), idSprites enables us to learn general transformations, thereby separating generalization from memorization and testing our main hypothesis. We hope that by releasing idSprites as a Python package we will encourage the research community to test their methods on our benchmark. Section 3.2 describes an implementation of our disentangled learning framework in the context of class-incremental continual learning. Our proof of concept consists of an equivariant network that learns to regress the parameters of an affine transform that maps any input into its canonical form, a normalisation module that applies the predicted affine transformation to the input, and an exemplar buffer that stores a single exemplar per class. Figure 1 shows the main components of our framework. Contributions We summarize the most important contributions of this work below: Contributions • We introduce a new framework for generating continual classification and disentanglement benchmarks that for the first time allows testing continual learning methods over thousands of tasks. We will open-source our software package upon acceptance. • We propose a novel continual learning paradigm based on learning symmetry transformations, which circumvents catastrophic forgetting by separating gradient-based model updates from explicit memory edits. • We demonstrate that as the number of tasks grows, regularization-based continual learning methods quickly break down and replay-based methods either deteriorate in performance or become impractical due to extensive use of memory and compute. • We show that our approach exhibits significant forward and backward transfer, strong open-set classification performance, and excellent zero-shot generalisation. It can learn over hundreds of tasks with a constant computational budget and a slowly growing memory footprint. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Disentangled Continual Learning: Separating Memory Edits from Model Updates

Author:
(1) V V Obukhov, Institute of research and development, Tomsk State Pedagogical University, Kievskaya str. 60, Tomsk 634061, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia (E-mail: obukhov@tspu.edu.ru);
(2) K E Osetrin, Center for Mathematical and Computer Physics, Tomsk State Pedagogical University, Kievskaya str. 60, Tomsk 634061, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia;
(3) A V Shapovalov, Department of Theoretical Physics, National Research Tomsk State University, Lenina pr. 36, Tomsk 634050, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia. Table of Links Abstract and Introduction Conclusion, Statements and Declarations, and Bibliography Conclusion Summarising, we believe that the statement made in the papers [1, 2] about new scientific results in the problem of complete separation of variables in the HamiltonJacobi equation for geodesics does not correspond to reality. We did not find any new results in comparison with those contained in the theorem on necessary and sufficient conditions of Vladimir N. Shapovalov. The results of the article in question, unlike the original ones, have omissions, and the article contains incorrect statements. Thus, the methods and results proposed in the works [1, 2] do not contain scientific novelty and represent a special case of the theory of complete separation of variables in the free Hamilton-Jacobi equation, finally constructed in the 70s of the last century. Statements and Declarations Data availability statement All necessary data and references to external sources are contained in the text of the manuscript. All information sources used in the work are publicly available and refer to open publications in scientific journals and textbooks. Compliance with Ethical Standards The authors declare no conflict of interest. Bibliography [1] Katanaev M O, Complete separation of variables in the geodesic HamiltonJacobi equation in four dimensions, 2023 Physica Scripta 98 104001 URL https://dx.doi.org/10.1088/1402-4896/acf251 [2] Katanaev M O 2023 Complete separation of variables in the geodesic Hamilton-Jacobi equation (Preprint 2305.02222) URL https://arxiv.org/abs/2305.02222 [3] Obukhov V, Hamilton–Jacobi equation for a charged test particle in the St¨ackel space of type (2.0), 2020 Symmetry 12 URL https://www.mdpi.com/2073-8994/12/8/1289 [4] Obukhov V, Separation of variables in Hamilton–Jacobi equation for a charged test particle in the St¨ackel spaces of type (2.1), 2020 International Journal of Geometric Methods in Modern Physics 17 2050186 URL https://doi.org/10.1142/S0219887820501868 [5] Obukhov V, Separation of variables in Hamilton–Jacobi and Klein–Gordon–Fock equations for a charged test particle in the St¨ackel spaces of type (1.1), 2021 International Journal of Geometric Methods in Modern Physics 18 2150036 URL https://doi.org/10.1142/S0219887821500365 [6] Obukhov V V 2006 St¨ackel spaces in gravity theory (Tomsk: Tomsk State Pedagogical University) ISBN 978-5-89428-242-8 [7] Bagrov V G, Meshkov A G, Shapovalov V N and Shapovalov A V, Separation of variables in the Klein-Gordon equations. I, 1973 Soviet Physics Journal 16 1533–1538 URL https://doi.org/10.1007/BF00889957 [8] Shapovalov V N, Symmetry and separation of variables in Hamilton-Jacobi equations. I, 1978 Soviet Physics Journal 21 1124–1129 URL https://doi.org/10.1007/BF00894559 [9] Shapovalov V N, St¨ackel spaces, 1979 Siberian Mathematical Journal 20 790–800 URL https://doi.org/10.1007/BF00971844 [10] Bagrov V G, Obukhov V V and Shapovalov A V, Special St¨ackel electrovac spacetimes, 1986 Pramana - J. Phys. 26 93–108 URL https://doi.org/10.1007/BF02847629 [11] Bagrov V G and Obukhov V V, Classes of exact solutions of the Einstein–Maxwell equations, 1983 Annalen der Physik 495 181–188 URL https://doi.org/10.1002/andp.19834950402 [12] Bagrov V G, Obukhov V V and Osetrin K E, Classification of null St¨ackel electrovac metrics with cosmological constant, 1988 General Relativity and Gravitation 20 1141–1154 URL https://doi.org/10.1007/BF00758935 This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. Author: (1) V V Obukhov, Institute of research and development, Tomsk State Pedagogical University, Kievskaya str. 60, Tomsk 634061, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia (E-mail: obukhov@tspu.edu.ru); (2) K E Osetrin, Center for Mathematical and Computer Physics, Tomsk State Pedagogical University, Kievskaya str. 60, Tomsk 634061, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia; (3) A V Shapovalov, Department of Theoretical Physics, National Research Tomsk State University, Lenina pr. 36, Tomsk 634050, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia. Author: Author: (1) V V Obukhov, Institute of research and development, Tomsk State Pedagogical University, Kievskaya str. 60, Tomsk 634061, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia (E-mail: obukhov@tspu.edu.ru); (2) K E Osetrin, Center for Mathematical and Computer Physics, Tomsk State Pedagogical University, Kievskaya str. 60, Tomsk 634061, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia; (3) A V Shapovalov, Department of Theoretical Physics, National Research Tomsk State University, Lenina pr. 36, Tomsk 634050, Russia & International laboratory of theoretical cosmology, Tomsk State University of Control Systems and Radioelectronics, Lenina pr. 40, Tomsk 634050, Russia. Table of Links Abstract and Introduction Abstract and Introduction Conclusion, Statements and Declarations, and Bibliography Conclusion, Statements and Declarations, and Bibliography Conclusion Summarising, we believe that the statement made in the papers [1, 2] about new scientific results in the problem of complete separation of variables in the HamiltonJacobi equation for geodesics does not correspond to reality. We did not find any new results in comparison with those contained in the theorem on necessary and sufficient conditions of Vladimir N. Shapovalov. The results of the article in question, unlike the original ones, have omissions, and the article contains incorrect statements. Thus, the methods and results proposed in the works [1, 2] do not contain scientific novelty and represent a special case of the theory of complete separation of variables in the free Hamilton-Jacobi equation, finally constructed in the 70s of the last century. Statements and Declarations Data availability statement Data availability statement All necessary data and references to external sources are contained in the text of the manuscript. All information sources used in the work are publicly available and refer to open publications in scientific journals and textbooks. Compliance with Ethical Standards Compliance with Ethical Standards The authors declare no conflict of interest. Bibliography [1] Katanaev M O, Complete separation of variables in the geodesic HamiltonJacobi equation in four dimensions, 2023 Physica Scripta 98 104001 URL https://dx.doi.org/10.1088/1402-4896/acf251 [2] Katanaev M O 2023 Complete separation of variables in the geodesic Hamilton-Jacobi equation (Preprint 2305.02222) URL https://arxiv.org/abs/2305.02222 [3] Obukhov V, Hamilton–Jacobi equation for a charged test particle in the St¨ackel space of type (2.0), 2020 Symmetry 12 URL https://www.mdpi.com/2073-8994/12/8/1289 [4] Obukhov V, Separation of variables in Hamilton–Jacobi equation for a charged test particle in the St¨ackel spaces of type (2.1), 2020 International Journal of Geometric Methods in Modern Physics 17 2050186 URL https://doi.org/10.1142/S0219887820501868 [5] Obukhov V, Separation of variables in Hamilton–Jacobi and Klein–Gordon–Fock equations for a charged test particle in the St¨ackel spaces of type (1.1), 2021 International Journal of Geometric Methods in Modern Physics 18 2150036 URL https://doi.org/10.1142/S0219887821500365 [6] Obukhov V V 2006 St¨ackel spaces in gravity theory (Tomsk: Tomsk State Pedagogical University) ISBN 978-5-89428-242-8 [7] Bagrov V G, Meshkov A G, Shapovalov V N and Shapovalov A V, Separation of variables in the Klein-Gordon equations. I, 1973 Soviet Physics Journal 16 1533–1538 URL https://doi.org/10.1007/BF00889957 [8] Shapovalov V N, Symmetry and separation of variables in Hamilton-Jacobi equations. I, 1978 Soviet Physics Journal 21 1124–1129 URL https://doi.org/10.1007/BF00894559 [9] Shapovalov V N, St¨ackel spaces, 1979 Siberian Mathematical Journal 20 790–800 URL https://doi.org/10.1007/BF00971844 [10] Bagrov V G, Obukhov V V and Shapovalov A V, Special St¨ackel electrovac spacetimes, 1986 Pramana - J. Phys. 26 93–108 URL https://doi.org/10.1007/BF02847629 [11] Bagrov V G and Obukhov V V, Classes of exact solutions of the Einstein–Maxwell equations, 1983 Annalen der Physik 495 181–188 URL https://doi.org/10.1002/andp.19834950402 [12] Bagrov V G, Obukhov V V and Osetrin K E, Classification of null St¨ackel electrovac metrics with cosmological constant, 1988 General Relativity and Gravitation 20 1141–1154 URL https://doi.org/10.1007/BF00758935 This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. available on arxiv

Revisiting Separation of Variables in Hamilton-Jacobi Equations: Conclusion, Statements and More

Revisiting Separation of Variables in Hamilton-Jacobi Equations: Abstract and Introduction

Greetings, fellow developers! Throughout my career, I've conducted over 100 interviews with developers of all levels, from junior devs to seasoned veterans. And you know what? In that time, I've crafted a set of 10 tricky questions that I pose to every new candidate. Why? Well, it's simple: these questions allow me to go beyond merely assessing someone's technical skills. They delve deeper, into their thought process, problem-solving approach, teamwork abilities, and other crucial qualities. Are you ready for this intellectual battle? Then brace yourselves, because these questions aren't for the faint of heart! 1. Preventing Concurrent Editing of Articles Suppose you are implementing a journal editing system where editors can edit articles in the admin panel. How would you prevent two editors from simultaneously editing the same article and overwriting each other's changes? 2. Principles of Zero Downtime Deployment What are the principles of zero downtime deployment, an approach that deploys an application without service downtime? How is this achieved? 3. Criticisms of Your Preferred Framework Name aspects of your preferred framework that you dislike or that don't work the way you'd like. Explain how and where these issues arise and how they could be improved. 4. Handling Long-Running Cron Jobs What could happen if your cron job, which runs every minute, starts taking more than one minute to execute? How can this be prevented? 5. Denormalizing Data in Databases Have you ever encountered the need to denormalize data in a database? If so, what problem were you trying to solve, what challenges did you face, and how could it have been done better? 6. Mocking External Calls in Tests If you write tests, how do you circumvent the issue of external calls made by the code you're testing? Consider the scenario where external calls are prohibited on CI (continuous integration). Why is this a good practice? 7. Synchronizing Email Changes with External Systems Suppose your system implements email change functionality. However, the email is also stored in an external system, such as a payment processor, which sends emails to users (but users don't directly interact with it). How would you implement email synchronization with the external system? 8. Discovering Production Errors How do you learn about errors that occur in production? Do you rely on user reports or have automated mechanisms in place? 9. Designing a Simple Chat Application How would you design a simple chat application? You can ask the backend engineer about the backend part and the frontend engineer about the frontend part. There could be many branches in this conversation. 10. Isolating Tests in a Multi-User Environment How is test isolation ensured when multiple tests run concurrently and access the database, potentially modifying it? If your framework doesn't provide isolation, how would you implement or improve it? Additional Tips: Encourage the candidate to elaborate on their thought process and reasoning behind their answers. Assess their problem-solving skills, ability to think critically, and knowledge of relevant technologies. Gauge their communication and collaboration skills throughout the interview. This article was inspired by twitter @mokevnin Greetings, fellow developers! Throughout my career, I've conducted over 100 interviews with developers of all levels, from junior devs to seasoned veterans. And you know what? In that time, I've crafted a set of 10 tricky questions that I pose to every new candidate. 100 interviews 10 tricky questions Why? Well, it's simple: these questions allow me to go beyond merely assessing someone's technical skills. They delve deeper, into their thought process, problem-solving approach, teamwork abilities, and other crucial qualities. Are you ready for this intellectual battle? Then brace yourselves, because these questions aren't for the faint of heart! 1. Preventing Concurrent Editing of Articles 1. Preventing Concurrent Editing of Articles Suppose you are implementing a journal editing system where editors can edit articles in the admin panel. How would you prevent two editors from simultaneously editing the same article and overwriting each other's changes? 2. Principles of Zero Downtime Deployment 2. Principles of Zero Downtime Deployment What are the principles of zero downtime deployment, an approach that deploys an application without service downtime? How is this achieved? 3. Criticisms of Your Preferred Framework 3. Criticisms of Your Preferred Framework Name aspects of your preferred framework that you dislike or that don't work the way you'd like. Explain how and where these issues arise and how they could be improved. 4. Handling Long-Running Cron Jobs 4. Handling Long-Running Cron Jobs What could happen if your cron job, which runs every minute, starts taking more than one minute to execute? How can this be prevented? 5. Denormalizing Data in Databases 5. Denormalizing Data in Databases Have you ever encountered the need to denormalize data in a database? If so, what problem were you trying to solve, what challenges did you face, and how could it have been done better? 6. Mocking External Calls in Tests 6. Mocking External Calls in Tests If you write tests, how do you circumvent the issue of external calls made by the code you're testing? Consider the scenario where external calls are prohibited on CI (continuous integration). Why is this a good practice? 7. Synchronizing Email Changes with External Systems 7. Synchronizing Email Changes with External Systems Suppose your system implements email change functionality. However, the email is also stored in an external system, such as a payment processor, which sends emails to users (but users don't directly interact with it). How would you implement email synchronization with the external system? 8. Discovering Production Errors 8. Discovering Production Errors How do you learn about errors that occur in production? Do you rely on user reports or have automated mechanisms in place? 9. Designing a Simple Chat Application 9. Designing a Simple Chat Application How would you design a simple chat application? You can ask the backend engineer about the backend part and the frontend engineer about the frontend part. There could be many branches in this conversation. 10. Isolating Tests in a Multi-User Environment 10. Isolating Tests in a Multi-User Environment How is test isolation ensured when multiple tests run concurrently and access the database, potentially modifying it? If your framework doesn't provide isolation, how would you implement or improve it? Additional Tips: Additional Tips: Encourage the candidate to elaborate on their thought process and reasoning behind their answers. Assess their problem-solving skills, ability to think critically, and knowledge of relevant technologies. Gauge their communication and collaboration skills throughout the interview. This article was inspired by twitter @mokevnin

These Coding Interview Questions Will Separate the Pros From the Amateurs

এই কোডিং ইন্টারভিউ প্রশ্ন অপেশাদারদের থেকে পেশাদারদের আলাদা করবে

Diese Fragen im Programmierinterview unterscheiden Profis von Amateuren

Estas preguntas de la entrevista de codificación separarán a los profesionales de los aficionados

Ces questions d'entretien de codage sépareront les pros des amateurs

ये कोडिंग साक्षात्कार प्रश्न पेशेवरों को शौकीनों से अलग कर देंगे

これらのコーディング面接の質問はプロとアマチュアを区別するでしょう

코딩 인터뷰 질문으로 전문가와 아마추어를 구분할 수 있습니다.

Estas perguntas da entrevista sobre codificação separarão os profissionais dos amadores

Эти вопросы на собеседовании по программированию отделят профессионалов от любителей

Bu Kodlama Mülakat Soruları Profesyonelleri Amatörlerden Ayıracak

Những câu hỏi phỏng vấn viết mã này sẽ tách biệt những người chuyên nghiệp khỏi những người nghiệp dư

这些编程面试问题将区分专业人士和业余人士

Explanation of ERC-6900: Modular Smart Contract Accounts and Plugins

Transforming Legacy with Domain-Driven Design, III: Tactic

How to Deploy a SPA on AWS With S3 and CloudFront: The Right Way to Do It

Primer on Large Language Model (LLM) Inference Optimizations: 2. Introduction to Artificial Intelligence (AI) Accelerators

Primer on Large Language Model (LLM) Inference Optimizations: 2. Úvod do akcelerátorů umělé inteligence (AI)

Introducción a las optimizaciones de inferencia de modelos de lenguaje grandes (LLM): 2. Introducción a los aceleradores de inteligencia artificial (IA)

Manual sobre as optimizacións de inferencia do modelo de linguaxe grande (LLM): 2. Introdución aos aceleradores de intelixencia artificial (IA)

वृहद भाषा मॉडल (एलएलएम) अनुमान अनुकूलन पर प्राइमर: 2. कृत्रिम बुद्धिमत्ता (एआई) त्वरक का परिचय

Introduzione alle ottimizzazioni dell'inferenza del Large Language Model (LLM): 2. Introduzione agli acceleratori dell'intelligenza artificiale (IA)

大規模言語モデル (LLM) 推論最適化入門: 2. 人工知能 (AI) アクセラレータの概要

Primer on Large Language Model (LLM) დასკვნის ოპტიმიზაცია: 2. შესავალი ხელოვნური ინტელექტის (AI) ამაჩქარებლებში

Үлкен тілдік үлгідегі (LLM) қорытындыны оңтайландырудағы праймер: 2. Жасанды интеллект (AI) үдеткіштеріне кіріспе

Том хэлний загвар (LLM) дүгнэлтийг оновчтой болгох праймер: 2. Хиймэл оюун ухаан (AI) хурдасгуурын танилцуулга

Horudhac ku saabsan Qaabka Luqadda Weyn (LLM) Kordhinta Farshaxanka: 2. Horudhaca Sirdoonka Artificial (AI) Dardariyeyaasha

Primer on Large Language Model (LLM) Inference Optimizations: 2. Panimula sa Artificial Intelligence (AI) Accelerators

Sách nhập môn về Tối ưu hóa suy luận mô hình ngôn ngữ lớn (LLM): 2. Giới thiệu về bộ tăng tốc trí tuệ nhân tạo (AI)

Don’t Panic! How Clear Communication and Strategy Can Prevent Cybersecurity Disasters

5 Useful Property Wrappers in Swift

Stories, News & Mentions about SEPANA

Sepana’s Stories on HackerNoon

Sepana's News & Mentions