Search Publications
Publications
2021 |
Danassis, Panayiotis; Wiedemair, Florian; Faltings, Boi PermLink Improving Multi-agent Coordination by Learning to Estimate Contention Conference Proceedings of the 30th International Joint Conference on Artificial Intelligence, IJCAI '21 International Joint Conferences on Artificial Intelligence Organization, 2021. @conference{danassis2021ijcai, title = {Improving Multi-agent Coordination by Learning to Estimate Contention}, author = {Panayiotis Danassis and Florian Wiedemair and Boi Faltings}, url = {https://lia.epfl.ch/wp-content/uploads/2021/06/ijcai2021_1.pdf}, year = {2021}, date = {2021-08-19}, booktitle = {Proceedings of the 30th International Joint Conference on Artificial Intelligence}, journal = {Proceedings of the 30th International Joint Conference on Artificial Intelligence}, publisher = {International Joint Conferences on Artificial Intelligence Organization}, series = {IJCAI '21}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Petrescu, Diana; Antognini, Diego; Faltings, Boi PermLink Multi-Step Critiquing User Interface for Recommender Systems Conference Demo Proceedings of the 15th ACM Conference on Recommender Systems Proceedings (RecSys 2021), 2021. @conference{petrescu_fastcritiquing_2021, title = {Multi-Step Critiquing User Interface for Recommender Systems}, author = {Diana Petrescu and Diego Antognini and Boi Faltings}, year = {2021}, date = {2021-07-30}, booktitle = {Demo Proceedings of the 15th ACM Conference on Recommender Systems Proceedings (RecSys 2021)}, abstract = {https://arxiv.org/pdf/2107.06416.pdf}, keywords = {}, pubstate = {published}, tppubtype = {conference} } https://arxiv.org/pdf/2107.06416.pdf |
Antognini, Diego; Faltings, Boi PermLink Fast Multi-Step Critiquing for VAE-based Recommender Systems Conference Proceedings of the 15th ACM Conference on Recommender Systems Proceedings (RecSys 2021), 2021. @conference{antognini_fastcritiquing_2021, title = {Fast Multi-Step Critiquing for VAE-based Recommender Systems}, author = {Diego Antognini and Boi Faltings}, url = {https://arxiv.org/pdf/2105.00774.pdf}, year = {2021}, date = {2021-07-06}, booktitle = {Proceedings of the 15th ACM Conference on Recommender Systems Proceedings (RecSys 2021)}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Padh, Kirtan; Antognini, Diego; Glaude, Emma Lejal; Faltings, Boi; Musat, Claudiu PermLink Addressing Fairness in Classificationwith a Model-Agnostic Multi-Objective Algorithm Conference Proceedings of the 37th Conference on Uncertainty in Artificial Intelligence (UAI 2021), 2021. @conference{kirtan2021, title = {Addressing Fairness in Classificationwith a Model-Agnostic Multi-Objective Algorithm}, author = {Kirtan Padh and Diego Antognini and Emma Lejal Glaude and Boi Faltings and Claudiu Musat}, url = {https://arxiv.org/pdf/2009.04441.pdf}, year = {2021}, date = {2021-05-12}, booktitle = {Proceedings of the 37th Conference on Uncertainty in Artificial Intelligence (UAI 2021)}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Antognini, Diego; Faltings, Boi PermLink Rationalization through Concepts Conference Findings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL 2021), 2021. @conference{antognini_concept_2021, title = {Rationalization through Concepts}, author = {Diego Antognini and Boi Faltings }, url = {https://arxiv.org/pdf/2105.04837.pdf}, year = {2021}, date = {2021-05-06}, booktitle = {Findings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL 2021)}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Danassis, Panayiotis; Erden, Zeki Doruk; Faltings, Boi PermLink Improved Cooperation by Exploiting a Common Signal Conference Proceedings of the 20th International Conference on Autonomous Agents and MultiAgent Systems, AAMAS '21 International Foundation for Autonomous Agents and Multiagent Systems, 2021. @conference{danassis2021aamas, title = {Improved Cooperation by Exploiting a Common Signal}, author = {Panayiotis Danassis and Zeki Doruk Erden and Boi Faltings}, url = {https://lia.epfl.ch/wp-content/uploads/2021/06/aamas2021.pdf}, year = {2021}, date = {2021-05-03}, booktitle = {Proceedings of the 20th International Conference on Autonomous Agents and MultiAgent Systems}, publisher = {International Foundation for Autonomous Agents and Multiagent Systems}, series = {AAMAS '21}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Antognini, Diego; Musat, Claudiu; Faltings, Boi PermLink Interacting with Explanations through Critiquing Conference Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI 2021), 2021. @conference{antognini22021, title = {Interacting with Explanations through Critiquing}, author = {Diego Antognini and Claudiu Musat and Boi Faltings }, url = {https://arxiv.org/pdf/2005.11067.pdf}, year = {2021}, date = {2021-04-01}, booktitle = {Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI 2021)}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Antognini, Diego; Musat, Claudiu; Faltings, Boi PermLink Multi-Dimensional Explanation of Target Variables from Documents Conference Proceedings of the 35th AAAI Conference on Artificial Intelligence (AAAI 2021), 2021. @conference{antognini2021, title = {Multi-Dimensional Explanation of Target Variables from Documents}, author = {Diego Antognini and Claudiu Musat and Boi Faltings }, editor = {The 35th AAAI Conference on Artificial Intelligence (AAAI), 2021}, url = {https://arxiv.org/pdf/1909.11386.pdf}, year = {2021}, date = {2021-02-01}, booktitle = {Proceedings of the 35th AAAI Conference on Artificial Intelligence (AAAI 2021)}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
2020 |
A. Richardson, Filos-Ratsikas A; Faltings, B PermLink Budget-Bounded Incentives for Federated Learning Book Chapter Q. Yang L. Fan, Yu H (Ed.): 12500 , pp. 176-188, Springer, 2020. @inbook{Richardson2020, title = {Budget-Bounded Incentives for Federated Learning}, author = {A. Richardson, A. Filos-Ratsikas and B. Faltings}, editor = {Q. Yang, L. Fan, H. Yu}, url = {https://lia.epfl.ch/wp-content/uploads/2021/02/BudgetBoundedIncentivesAug18-2.pdf}, year = {2020}, date = {2020-11-26}, volume = {12500}, pages = {176-188}, publisher = {Springer}, series = {LNCS}, abstract = {We consider federated learning settings with independent, self-interested participants. As all contributions are made privately, participants may be tempted to free-ride and provide redundant or low-quality data while still enjoying the benefits of the FL model. In Federated Learning, this is especially harmful as low-quality data can degrade the quality of the FL model. Free-riding can be countered by giving incentives to participants to provide truthful data. While there are game-theoretic schemes for rewarding truthful data, they do not take into account redundancy of data with previous contributions. This creates arbitrage opportunities where participants can gain rewards for redundant data, and the federation may be forced to pay out more incentives than justified by the value of the FL model. We show how a scheme based on influence can both guarantee that the incentive budget is bounded in proportion to the value of the FL model, and that truthfully reporting data is the dominant strategy of the participants. We show that under reasonable conditions, this result holds even when the testing data is provided by participants.}, keywords = {}, pubstate = {published}, tppubtype = {inbook} } We consider federated learning settings with independent, self-interested participants. As all contributions are made privately, participants may be tempted to free-ride and provide redundant or low-quality data while still enjoying the benefits of the FL model. In Federated Learning, this is especially harmful as low-quality data can degrade the quality of the FL model. Free-riding can be countered by giving incentives to participants to provide truthful data. While there are game-theoretic schemes for rewarding truthful data, they do not take into account redundancy of data with previous contributions. This creates arbitrage opportunities where participants can gain rewards for redundant data, and the federation may be forced to pay out more incentives than justified by the value of the FL model. We show how a scheme based on influence can both guarantee that the incentive budget is bounded in proportion to the value of the FL model, and that truthfully reporting data is the dominant strategy of the participants. We show that under reasonable conditions, this result holds even when the testing data is provided by participants. |
Fei Mi Xiaoyu Lin, Boi Faltings PermLink The ACM Conference on Recommender Systems (RecSys), 2020, (RecSys 2020 best short paper award). @conference{Miii2020, title = {ADER: Adaptively Distilled Exemplar Replay Towards Continual Learning for Session-based Recommendation}, author = {Fei Mi, Xiaoyu Lin, Boi Faltings}, url = {https://dl.acm.org/doi/abs/10.1145/3383313.3412218}, year = {2020}, date = {2020-09-07}, booktitle = {The ACM Conference on Recommender Systems (RecSys)}, pages = {408-413}, note = {RecSys 2020 best short paper award}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Mi, Fei; Faltings, Boi PermLink Memory Augmented Neural Model for Incremental Session-based Recommendation Conference Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI 2020), 2020. @conference{Mi2020, title = {Memory Augmented Neural Model for Incremental Session-based Recommendation}, author = {Fei Mi and Boi Faltings}, url = {https://www.ijcai.org/Proceedings/2020/0300.pdf }, year = {2020}, date = {2020-08-08}, booktitle = {Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI 2020)}, pages = {2169-2176}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Fei Mi Liangwei Chen, Mengjie Zhao Minlie Huang Boi Faltings PermLink Continual Learning for Natural Language Generation in Task-oriented Dialog Systems Conference Findings of the Association for Computational Linguistics: EMNLP 2020, 2020. @conference{Miiii2020, title = {Continual Learning for Natural Language Generation in Task-oriented Dialog Systems}, author = {Fei Mi, Liangwei Chen, Mengjie Zhao, Minlie Huang, Boi Faltings}, url = {https://www.aclweb.org/anthology/2020.findings-emnlp.310}, doi = {10.18653/v1/2020.findings-emnlp.310}, year = {2020}, date = {2020-08-08}, booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2020}, pages = {3461–3474}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Goel, Naman; Filos-Ratsikas, Aris; Faltings, Boi PermLink Peer-Prediction in the Presence of Outcome Dependent Lying Incentives Conference Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI 2020), 2020. @conference{Goel2020b, title = {Peer-Prediction in the Presence of Outcome Dependent Lying Incentives}, author = {Naman Goel and Aris Filos-Ratsikas and Boi Faltings}, url = {https://www.ijcai.org/Proceedings/2020/0018.pdf}, year = {2020}, date = {2020-07-11}, booktitle = {Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI 2020)}, pages = {124-131}, abstract = {We derive conditions under which a peer-consistency mechanism can be used to elicit truthful data from non-trusted rational agents when an aggregate statistic of the collected data affects the amount of their incentives to lie. Furthermore, we discuss the relative saving that can be achieved by the mechanism, compared to the rational outcome, if no such mechanism was implemented. Our work is motivated by distributed platforms, where decentralized data oracles collect information about real-world events, based on the aggregate information provided by often self-interested participants. We compare our theoretical observations with numerical simulations on two public real datasets.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } We derive conditions under which a peer-consistency mechanism can be used to elicit truthful data from non-trusted rational agents when an aggregate statistic of the collected data affects the amount of their incentives to lie. Furthermore, we discuss the relative saving that can be achieved by the mechanism, compared to the rational outcome, if no such mechanism was implemented. Our work is motivated by distributed platforms, where decentralized data oracles collect information about real-world events, based on the aggregate information provided by often self-interested participants. We compare our theoretical observations with numerical simulations on two public real datasets. |
Goel*, Naman; van Schreven*, Cyril; Filos-Ratsikas, Aris; Faltings, Boi PermLink Infochain: A Decentralized, Trustless and Transparent Oracle on Blockchain Conference Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI 2020), 2020. @conference{Goel2020c, title = {Infochain: A Decentralized, Trustless and Transparent Oracle on Blockchain}, author = {Naman Goel* and Cyril van Schreven* and Aris Filos-Ratsikas and Boi Faltings}, url = {https://www.ijcai.org/Proceedings/2020/0635.pdf}, doi = {https://doi.org/10.24963/ijcai.2020/635}, year = {2020}, date = {2020-07-11}, booktitle = {Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI 2020)}, pages = {4604-4610}, abstract = {Blockchain based systems allow various kinds of financial transactions to be executed in a decentralized manner. However, these systems often rely on a trusted third party (oracle) to get correct information about the real-world events, which trigger the financial transactions. In this paper, we identify two biggest challenges in building decentralized, trustless and transparent oracles. The first challenge is acquiring correct information about the real-world events without relying on a trusted information provider. We show how a peer-consistency incentive mechanism can be used to acquire truthful information from an untrusted and self-interested crowd, even when the crowd has outside incentives to provide wrong informations. The second is a system design and implementation challenge. For the first time, we show how to implement a trustless and transparent oracle in Ethereum. We discuss various non-trivial issues that arise in implementing peer-consistency mechanisms in Ethereum, suggest several optimizations to reduce gas cost and provide empirical analysis.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Blockchain based systems allow various kinds of financial transactions to be executed in a decentralized manner. However, these systems often rely on a trusted third party (oracle) to get correct information about the real-world events, which trigger the financial transactions. In this paper, we identify two biggest challenges in building decentralized, trustless and transparent oracles. The first challenge is acquiring correct information about the real-world events without relying on a trusted information provider. We show how a peer-consistency incentive mechanism can be used to acquire truthful information from an untrusted and self-interested crowd, even when the crowd has outside incentives to provide wrong informations. The second is a system design and implementation challenge. For the first time, we show how to implement a trustless and transparent oracle in Ethereum. We discuss various non-trivial issues that arise in implementing peer-consistency mechanisms in Ethereum, suggest several optimizations to reduce gas cost and provide empirical analysis. |
Goel, Naman; Rutagarama, Maxime; Faltings, Boi PermLink Tackling Peer to Peer Discrimination in the Sharing Economy Conference Proceedings of the 12th ACM Web Science Conference (WebSci 2020), ACM, 2020, (Best paper award). @conference{Goel2020, title = {Tackling Peer to Peer Discrimination in the Sharing Economy}, author = {Naman Goel and Maxime Rutagarama and Boi Faltings}, url = {https://lia.epfl.ch/~goel/upload/doc/papers/2020/goel_websci20.pdf}, doi = {https://doi.org/10.1145/3394231.3397926}, year = {2020}, date = {2020-07-07}, booktitle = {Proceedings of the 12th ACM Web Science Conference (WebSci 2020)}, pages = {355-361}, publisher = {ACM}, abstract = {Sharing economy platforms such as Airbnb and Uber face a major challenge in the form of peer-to-peer discrimination based on sensitive personal attributes such as race and gender. As shown by a recent study under controlled settings, reputation systems can eliminate social biases on these platforms by building trust between the users. However, for this to work in practice, the reputation systems must themselves be non-discriminatory. In fact, a biased reputation system will further reinforce the bias and create a vicious feedback loop. Given that the reputation scores are generally aggregates of ratings provided by human users to one another, it is not surprising that the scores often inherit the human bias. In this paper, we address the problem of making reputation systems on sharing economy platforms more fair and unbiased. We show that a game-theoretical incentive mechanism can be used to encourage users to go against common bias and provide a truthful rating about others, obtained through a more careful and deeper evaluation. In situations where an incentive mechanism can’t be implemented, we show that a simple post-processing approach can also be used to correct bias in the reputation scores, while minimizing the loss in the useful information provided by the scores. We evaluate the proposed solution on synthetic and real datasets from Airbnb.}, note = {Best paper award}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Sharing economy platforms such as Airbnb and Uber face a major challenge in the form of peer-to-peer discrimination based on sensitive personal attributes such as race and gender. As shown by a recent study under controlled settings, reputation systems can eliminate social biases on these platforms by building trust between the users. However, for this to work in practice, the reputation systems must themselves be non-discriminatory. In fact, a biased reputation system will further reinforce the bias and create a vicious feedback loop. Given that the reputation scores are generally aggregates of ratings provided by human users to one another, it is not surprising that the scores often inherit the human bias. In this paper, we address the problem of making reputation systems on sharing economy platforms more fair and unbiased. We show that a game-theoretical incentive mechanism can be used to encourage users to go against common bias and provide a truthful rating about others, obtained through a more careful and deeper evaluation. In situations where an incentive mechanism can’t be implemented, we show that a simple post-processing approach can also be used to correct bias in the reputation scores, while minimizing the loss in the useful information provided by the scores. We evaluate the proposed solution on synthetic and real datasets from Airbnb. |
Chatzopoulos, Dimitris; Gujar, Sujit; Faltings, Boi; Hui, Pan PermLink Mneme: A Mobile Distributed Ledger Conference IEEE Infocom 2020, IEEE, 2020. @conference{Chatzopoulos2020a, title = {Mneme: A Mobile Distributed Ledger}, author = {Dimitris Chatzopoulos and Sujit Gujar and Boi Faltings and Pan Hui}, url = {https://lia.epfl.ch/wp-content/uploads/2020/05/main.pdf}, year = {2020}, date = {2020-07-06}, booktitle = {IEEE Infocom 2020}, pages = {1897-1906}, publisher = {IEEE}, abstract = {Advances in mobile computing have paved the way for new types of distributed applications that can be executed solely by mobile devices on device-to-device (D2D) ecosystems (e.g., crowdsensing). More sophisticated applications, like cryptocurrencies, need distributed ledgers to function. Distributed ledgers, such as blockchains and directed acyclic graphs (DAGs), employ consensus protocols to add data in the form of blocks. However such protocols are designed for resourceful devices that are interconnected via the Internet. Moreover, existing distributed ledgers are not deployable to D2D ecosystems since their storage needs are continuously increasing. In this work, we introduce Mneme, a DAG-based distributed ledger that can be maintained solely by mobile devices and operates via two consensus protocols: Proof-of-Context (PoC) and Proof-of-Equivalence (PoE). PoC employs users’ context to add data on Mneme. PoE is executed periodically to summarize data and produce equivalent blocks that require less storage. We analyze the security of Mneme and justify the ability of PoC and PoE to guarantee the characteristics of distributed ledgers: persistence and liveness. Furthermore, we analyze potential attacks from malicious users and prove that the probability of a successful attack is inversely proportional to the square of the number of mobile users who maintain Mneme.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Advances in mobile computing have paved the way for new types of distributed applications that can be executed solely by mobile devices on device-to-device (D2D) ecosystems (e.g., crowdsensing). More sophisticated applications, like cryptocurrencies, need distributed ledgers to function. Distributed ledgers, such as blockchains and directed acyclic graphs (DAGs), employ consensus protocols to add data in the form of blocks. However such protocols are designed for resourceful devices that are interconnected via the Internet. Moreover, existing distributed ledgers are not deployable to D2D ecosystems since their storage needs are continuously increasing. In this work, we introduce Mneme, a DAG-based distributed ledger that can be maintained solely by mobile devices and operates via two consensus protocols: Proof-of-Context (PoC) and Proof-of-Equivalence (PoE). PoC employs users’ context to add data on Mneme. PoE is executed periodically to summarize data and produce equivalent blocks that require less storage. We analyze the security of Mneme and justify the ability of PoC and PoE to guarantee the characteristics of distributed ledgers: persistence and liveness. Furthermore, we analyze potential attacks from malicious users and prove that the probability of a successful attack is inversely proportional to the square of the number of mobile users who maintain Mneme. |
Mi, Fei; Kong, Lingjing; Lin, Tao; Yu, Kaicheng; Faltings, Boi PermLink Generalized Class Incremental Learning Workshop CVPR2020 Continual Learning Workshop, 2020. @workshop{Mii2020, title = {Generalized Class Incremental Learning}, author = {Fei Mi and Lingjing Kong and Tao Lin and Kaicheng Yu and Boi Faltings}, url = {https://openaccess.thecvf.com/content_CVPRW_2020/papers/w15/Mi_Generalized_Class_Incremental_Learning_CVPRW_2020_paper.pdf}, year = {2020}, date = {2020-06-14}, booktitle = {CVPR2020 Continual Learning Workshop}, keywords = {}, pubstate = {published}, tppubtype = {workshop} } |
Triastcyn, Aleksei; Faltings, Boi PermLink Federated Generative Privacy Journal Article IEEE Intelligent Systems, 35 (4), pp. 50-57, 2020, ISSN: 1541-1672. @article{triastcyn2020fedgp, title = {Federated Generative Privacy}, author = {Aleksei Triastcyn and Boi Faltings}, url = {https://lia.epfl.ch/wp-content/uploads/2020/05/fedgp_ieee_is_2020.pdf}, doi = {10.1109/MIS.2020.2993966}, issn = {1541-1672}, year = {2020}, date = {2020-05-12}, journal = {IEEE Intelligent Systems}, volume = {35}, number = {4}, pages = {50-57}, abstract = {We propose FedGP, a framework for privacy-preserving data release in the federated learning setting. We use generative adversarial networks, generator components of which are trained by FedAvg algorithm, to draw private artificial data samples and empirically assess the risk of information disclosure. Our experiments show that FedGP is able to generate labelled data of high quality to successfully train and validate supervised models. Finally, we demonstrate that our approach significantly reduces vulnerability of such models to model inversion attacks.}, keywords = {}, pubstate = {published}, tppubtype = {article} } We propose FedGP, a framework for privacy-preserving data release in the federated learning setting. We use generative adversarial networks, generator components of which are trained by FedAvg algorithm, to draw private artificial data samples and empirically assess the risk of information disclosure. Our experiments show that FedGP is able to generate labelled data of high quality to successfully train and validate supervised models. Finally, we demonstrate that our approach significantly reduces vulnerability of such models to model inversion attacks. |
Antognini, Diego; Faltings, Boi PermLink HotelRec: a Novel Very Large-Scale Hotel Recommendation Dataset Conference LREC 2020, Twelveth International Conference on Language Resources and Evaluation, 2020. @conference{antognini-faltings-2020-hotelrec, title = {HotelRec: a Novel Very Large-Scale Hotel Recommendation Dataset}, author = {Diego Antognini and Boi Faltings}, url = {https://lia.epfl.ch/wp-content/uploads/2020/02/LREC_2020_HotelRec.pdf}, year = {2020}, date = {2020-02-17}, booktitle = {LREC 2020, Twelveth International Conference on Language Resources and Evaluation}, pages = {4917-4923}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Antognini, Diego; Faltings, Boi PermLink GameWikiSum: a Novel Large Multi-Document Summarization Dataset Conference LREC 2020, Twelveth International Conference on Language Resources and Evaluation, (6645-6650), 2020. @conference{antognini-faltings-2020-gamewikisum, title = {GameWikiSum: a Novel Large Multi-Document Summarization Dataset}, author = {Diego Antognini and Boi Faltings}, url = {https://lia.epfl.ch/wp-content/uploads/2020/02/LREC_2020_GameWikiSum.pdf}, year = {2020}, date = {2020-02-17}, booktitle = {LREC 2020, Twelveth International Conference on Language Resources and Evaluation}, number = {6645-6650}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Milojkovic, Nikola; Antognini, Diego; Bergamin, Giancarlo; Faltings, Boi; Musat, Claudiu PermLink Multi-Gradient Descent for Multi-Objective Recommender Systems Proceeding Proceedings of the The Thirty-Fourth AAAI Conference on Artificial Intelligence 2020 - Workshop on Interactive and Conversational Recommendation Systems (WICRS), 2020. @proceedings{mikolovic2019, title = {Multi-Gradient Descent for Multi-Objective Recommender Systems}, author = {Nikola Milojkovic and Diego Antognini and Giancarlo Bergamin and Boi Faltings and Claudiu Musat}, url = {https://lia.epfl.ch/wp-content/uploads/2019/12/multi_obj_2019.pdf https://arxiv.org/abs/2001.00846}, year = {2020}, date = {2020-02-09}, publisher = {Proceedings of the The Thirty-Fourth AAAI Conference on Artificial Intelligence 2020 - Workshop on Interactive and Conversational Recommendation Systems (WICRS)}, abstract = {Recommender systems need to mirror the complexity of the environment they are applied in. The more we know about what might benefit the user, the more objectives the recommender system has. In addition there may be multiple stakeholders - sellers, buyers, shareholders - in addition to legal and ethical constraints. Simultaneously optimizing for a multitude of objectives, correlated and not correlated, having the same scale or not, has proven difficult so far. We introduce a stochastic multi-gradient descent approach to recommender systems (MGDRec) to solve this problem. We show that this exceeds state-of-the-art methods in traditional objective mixtures, like revenue and recall. Not only that, but through gradient normalization we can combine fundamentally different objectives, having diverse scales, into a single coherent framework. We show that uncorrelated objectives, like the proportion of quality products, can be improved alongside accuracy. Through the use of stochasticity, we avoid the pitfalls of calculating full gradients and provide a clear setting for its applicability.}, keywords = {}, pubstate = {published}, tppubtype = {proceedings} } Recommender systems need to mirror the complexity of the environment they are applied in. The more we know about what might benefit the user, the more objectives the recommender system has. In addition there may be multiple stakeholders - sellers, buyers, shareholders - in addition to legal and ethical constraints. Simultaneously optimizing for a multitude of objectives, correlated and not correlated, having the same scale or not, has proven difficult so far. We introduce a stochastic multi-gradient descent approach to recommender systems (MGDRec) to solve this problem. We show that this exceeds state-of-the-art methods in traditional objective mixtures, like revenue and recall. Not only that, but through gradient normalization we can combine fundamentally different objectives, having diverse scales, into a single coherent framework. We show that uncorrelated objectives, like the proportion of quality products, can be improved alongside accuracy. Through the use of stochasticity, we avoid the pitfalls of calculating full gradients and provide a clear setting for its applicability. |
Triastcyn, Aleksei; Faltings, Boi PermLink Bayesian Differential Privacy for Machine Learning Workshop AAAI Workshop on Privacy-Preserving Artificial Intelligence, 2020. @workshop{triastcyn2020bayesiandp, title = {Bayesian Differential Privacy for Machine Learning}, author = {Aleksei Triastcyn and Boi Faltings}, url = {https://www2.isye.gatech.edu/~fferdinando3/cfp/PPAI20/papers/paper_19.pdf}, year = {2020}, date = {2020-02-07}, booktitle = {AAAI Workshop on Privacy-Preserving Artificial Intelligence}, abstract = {We propose Bayesian differential privacy, a relaxation of differential privacy that provides sharper privacy guarantees for similarly distributed data, especially in difficult scenarios, such as deep learning. We derive a general privacy accounting method for iterative learning algorithms under Bayesian differential privacy and show that it is a generalisation of the well-known moments accountant. Our experiments show significant improvements in privacy guarantees for typical cases in deep learning datasets, such as MNIST and CIFAR-10, in some cases bringing the privacy budget ε from 8 down to 0.5.}, keywords = {}, pubstate = {published}, tppubtype = {workshop} } We propose Bayesian differential privacy, a relaxation of differential privacy that provides sharper privacy guarantees for similarly distributed data, especially in difficult scenarios, such as deep learning. We derive a general privacy accounting method for iterative learning algorithms under Bayesian differential privacy and show that it is a generalisation of the well-known moments accountant. Our experiments show significant improvements in privacy guarantees for typical cases in deep learning datasets, such as MNIST and CIFAR-10, in some cases bringing the privacy budget ε from 8 down to 0.5. |
2019 |
Aris Filos-Ratsikas, Yiannis Giannakopoulos ; Lazos, Philip PermLink The Pareto Frontier of Inefficiency in Mechanism Design Conference The 15th Conference on Web and Internet Economics (WINE 2019)., Springer, 2019. @conference{Filos-Ratsikas2019b, title = {The Pareto Frontier of Inefficiency in Mechanism Design}, author = {Aris Filos-Ratsikas, Yiannis Giannakopoulos and Philip Lazos}, url = {https://arxiv.org/abs/1809.03454}, year = {2019}, date = {2019-12-10}, booktitle = {The 15th Conference on Web and Internet Economics (WINE 2019).}, pages = {186-199}, publisher = {Springer}, abstract = {We study the trade-off between the Price of Anarchy (PoA) and the Price of Stability (PoS) in mechanism design, in the prototypical problem of unrelated machine scheduling. We give bounds on the space of feasible mechanisms with respect to the above metrics, and observe that two fundamental mechanisms, namely the First-Price (FP) and the Second-Price (SP), lie on the two opposite extrema of this boundary. Furthermore, for the natural class of anonymous task-independent mechanisms, we completely characterize the PoA/PoS Pareto frontier; we design a class of optimal mechanisms SPα that lie exactly on this frontier. In particular, these mechanisms range smoothly, with respect to parameter α≥1 across the frontier, between the First-Price (SP1) and Second-Price (SP∞) mechanisms. En route to these results, we also provide a definitive answer to an important question related to the scheduling problem, namely whether non-truthful mechanisms can provide better makespan guarantees in the equilibrium, compared to truthful ones. We answer this question in the negative, by proving that the Price of Anarchy of all scheduling mechanisms is at least n, where n is the number of machines.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } We study the trade-off between the Price of Anarchy (PoA) and the Price of Stability (PoS) in mechanism design, in the prototypical problem of unrelated machine scheduling. We give bounds on the space of feasible mechanisms with respect to the above metrics, and observe that two fundamental mechanisms, namely the First-Price (FP) and the Second-Price (SP), lie on the two opposite extrema of this boundary. Furthermore, for the natural class of anonymous task-independent mechanisms, we completely characterize the PoA/PoS Pareto frontier; we design a class of optimal mechanisms SPα that lie exactly on this frontier. In particular, these mechanisms range smoothly, with respect to parameter α≥1 across the frontier, between the First-Price (SP1) and Second-Price (SP∞) mechanisms. En route to these results, we also provide a definitive answer to an important question related to the scheduling problem, namely whether non-truthful mechanisms can provide better makespan guarantees in the equilibrium, compared to truthful ones. We answer this question in the negative, by proving that the Price of Anarchy of all scheduling mechanisms is at least n, where n is the number of machines. |
Triastcyn, Aleksei; Faltings, Boi PermLink Federated Learning with Bayesian Differential Privacy Inproceedings 2019 IEEE International Conference on Big Data (Big Data), pp. 2587-2596, IEEE, 2019, ISBN: 978-1-7281-0858-2. @inproceedings{triastcyn2019federatedbdp, title = {Federated Learning with Bayesian Differential Privacy}, author = {Aleksei Triastcyn and Boi Faltings}, doi = {10.1109/BigData47090.2019.9005465}, isbn = {978-1-7281-0858-2}, year = {2019}, date = {2019-12-09}, booktitle = {2019 IEEE International Conference on Big Data (Big Data)}, pages = {2587-2596}, publisher = {IEEE}, abstract = {We consider the problem of reinforcing federated learning with formal privacy guarantees. We propose to employ Bayesian differential privacy, a relaxation of differential privacy for similarly distributed data, to provide sharper privacy loss bounds. We adapt the Bayesian privacy accounting method to the federated setting and suggest multiple improvements for more efficient privacy budgeting at different levels. Our experiments show significant advantage over the state-of-the-art differential privacy bounds for federated learning on image classification tasks, including a medical application, bringing the privacy budget below ε = 1 at the client level, and below ε = 0.1 at the instance level. Lower amounts of noise also benefit the model accuracy and reduce the number of communication rounds.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We consider the problem of reinforcing federated learning with formal privacy guarantees. We propose to employ Bayesian differential privacy, a relaxation of differential privacy for similarly distributed data, to provide sharper privacy loss bounds. We adapt the Bayesian privacy accounting method to the federated setting and suggest multiple improvements for more efficient privacy budgeting at different levels. Our experiments show significant advantage over the state-of-the-art differential privacy bounds for federated learning on image classification tasks, including a medical application, bringing the privacy budget below ε = 1 at the client level, and below ε = 0.1 at the instance level. Lower amounts of noise also benefit the model accuracy and reduce the number of communication rounds. |
Antognini, Diego; Faltings, Boi PermLink Learning to Create Sentence Semantic Relation Graphs for Multi-Document Summarization Proceeding Proceedings of the Empirical Methods in Natural Language Processing 2019 - Workshop on New Frontiers in Summarization, 2019. @proceedings{antognini-faltings-2019-learning, title = {Learning to Create Sentence Semantic Relation Graphs for Multi-Document Summarization}, author = {Diego Antognini and Boi Faltings}, url = {https://lia.epfl.ch/wp-content/uploads/2019/11/D19-5404.pdf https://www.aclweb.org/anthology/D19-5404}, doi = {10.18653/v1/D19-5404}, year = {2019}, date = {2019-11-01}, booktitle = {Proceedings of the 2nd Workshop on New Frontiers in Summarization }, pages = {32--41}, address = {Proceedings of the Empirical Methods in Natural Language Processing 2019 - Workshop on New Frontiers in Summarization}, abstract = {Linking facts across documents is a challenging task, as the language used to express the same information in a sentence can vary significantly, which complicates the task of multi-document summarization. Consequently, existing approaches heavily rely on hand-crafted features, which are domain-dependent and hard to craft, or additional annotated data, which is costly to gather. To overcome these limitations, we present a novel method, which makes use of two types of sentence embeddings: universal embeddings, which are trained on a large unrelated corpus, and domain-specific embeddings, which are learned during training. To this end, we develop SemSentSum, a fully data-driven model able to leverage both types of sentence embeddings by building a sentence semantic relation graph. SemSentSum achieves competitive results on two types of summary, consisting of 665 bytes and 100 words. Unlike other state-of-the-art models, neither hand-crafted features nor additional annotated data are necessary, and the method is easily adaptable for other tasks. To our knowledge, we are the first to use multiple sentence embeddings for the task of multi-document summarization.}, keywords = {}, pubstate = {published}, tppubtype = {proceedings} } Linking facts across documents is a challenging task, as the language used to express the same information in a sentence can vary significantly, which complicates the task of multi-document summarization. Consequently, existing approaches heavily rely on hand-crafted features, which are domain-dependent and hard to craft, or additional annotated data, which is costly to gather. To overcome these limitations, we present a novel method, which makes use of two types of sentence embeddings: universal embeddings, which are trained on a large unrelated corpus, and domain-specific embeddings, which are learned during training. To this end, we develop SemSentSum, a fully data-driven model able to leverage both types of sentence embeddings by building a sentence semantic relation graph. SemSentSum achieves competitive results on two types of summary, consisting of 665 bytes and 100 words. Unlike other state-of-the-art models, neither hand-crafted features nor additional annotated data are necessary, and the method is easily adaptable for other tasks. To our knowledge, we are the first to use multiple sentence embeddings for the task of multi-document summarization. |