arXiv subjects

Anne-Marie Kermarrec

Publications and source records attributed to Anne-Marie Kermarrec.

At least 55 records · Page 3Linked to original sources

arXiv2023

Epidemic Learning: Boosting Decentralized Learning with Randomized Communication

We present Epidemic Learning (EL), a simple yet powerful decentralized learning (DL) algorithm that leverages changing communication topologies to achieve faster model convergence compared to conventional DL approaches. At each round of EL, each node sends its model updates to a random sample of $s$ other nodes (in a system of $n$ nodes). We provide an extensive theoretical analysis of EL, demonstrating that its changing topology culminates in superior convergence properties compared to the state-of-the-art (static and dynamic) topologies. Considering smooth non-convex loss functions, the number of transient iterations for EL, i.e., the rounds required to achieve asymptotic linear speedup, is in $O(n^3/s^2)$ which outperforms the best-known bound $O(n^3)$ by a factor of $s^2$, indicating the benefit of randomized communication for DL. We empirically evaluate EL in a 96-node network and compare its performance with state-of-the-art DL approaches. Our results illustrate that EL converges up to $ 1.7\times$ quicker than baseline DL algorithms and attains $2.2 $\% higher accuracy for the same communication volume.

cs.LG↗

arXiv2023

Decentralized Learning Made Easy with DecentralizePy

Decentralized learning (DL) has gained prominence for its potential benefits in terms of scalability, privacy, and fault tolerance. It consists of many nodes that coordinate without a central server and exchange millions of parameters in the inherently iterative process of machine learning (ML) training. In addition, these nodes are connected in complex and potentially dynamic topologies. Assessing the intricate dynamics of such networks is clearly not an easy task. Often in literature, researchers resort to simulated environments that do not scale and fail to capture practical and crucial behaviors, including the ones associated to parallelism, data transfer, network delays, and wall-clock time. In this paper, we propose DecentralizePy, a distributed framework for decentralized ML, which allows for the emulation of large-scale learning networks in arbitrary topologies. We demonstrate the capabilities of DecentralizePy by deploying techniques such as sparsification and secure aggregation on top of several topologies, including dynamic networks with more than one thousand nodes.

cs.DC↗

arXiv2022

Refined Convergence and Topology Learning for Decentralized SGD with Heterogeneous Data

One of the key challenges in decentralized and federated learning is to design algorithms that efficiently deal with highly heterogeneous data distributions across agents. In this paper, we revisit the analysis of the popular Decentralized Stochastic Gradient Descent algorithm (D-SGD) under data heterogeneity. We exhibit the key role played by a new quantity, called neighborhood heterogeneity, on the convergence rate of D-SGD. By coupling the communication topology and the heterogeneity, our analysis sheds light on the poorly understood interplay between these two concepts. We then argue that neighborhood heterogeneity provides a natural criterion to learn data-dependent topologies that reduce (and can even eliminate) the otherwise detrimental effect of data heterogeneity on the convergence time of D-SGD. For the important case of classification with label skew, we formulate the problem of learning such a good topology as a tractable optimization problem that we solve with a Frank-Wolfe algorithm. As illustrated over a set of simulated and real-world experiments, our approach provides a principled way to design a sparse topology that balances the convergence speed and the per-iteration communication costs of D-SGD under data heterogeneity.

cs.LG↗

arXiv2022

TEE-based decentralized recommender systems: The raw data sharing redemption

Recommenders are central in many applications today. The most effective recommendation schemes, such as those based on collaborative filtering (CF), exploit similarities between user profiles to make recommendations, but potentially expose private data. Federated learning and decentralized learning systems address this by letting the data stay on user's machines to preserve privacy: each user performs the training on local data and only the model parameters are shared. However, sharing the model parameters across the network may still yield privacy breaches. In this paper, we present REX, the first enclave-based decentralized CF recommender. REX exploits Trusted execution environments (TEE), such as Intel software guard extensions (SGX), that provide shielded environments within the processor to improve convergence while preserving privacy. Firstly, REX enables raw data sharing, which ultimately speeds up convergence and reduces the network load. Secondly, REX fully preserves privacy. We analyze the impact of raw data sharing in both deep neural network (DNN) and matrix factorization (MF) recommenders and showcase the benefits of trusted environments in a full-fledged implementation of REX. Our experimental results demonstrate that through raw data sharing, REX significantly decreases the training time by 18.3x and the network load by 2 orders of magnitude over standard decentralized approaches that share only parameters, while fully protecting privacy by leveraging trustworthy hardware enclaves with very little overhead.

cs.DC↗

arXiv2021

D-Cliques: Compensating for Data Heterogeneity with Topology in Decentralized Federated Learning

The convergence speed of machine learning models trained with Federated Learning is significantly affected by heterogeneous data partitions, even more so in a fully decentralized setting without a central server. In this paper, we show that the impact of label distribution skew, an important type of data heterogeneity, can be significantly reduced by carefully designing the underlying communication topology. We present D-Cliques, a novel topology that reduces gradient bias by grouping nodes in sparsely interconnected cliques such that the label distribution in a clique is representative of the global label distribution. We also show how to adapt the updates of decentralized SGD to obtain unbiased gradients and implement an effective momentum with D-Cliques. Our extensive empirical evaluation on MNIST and CIFAR10 demonstrates that our approach provides similar convergence speed as a fully-connected topology, which provides the best convergence in a data heterogeneous setting, with a significant reduction in the number of edges and messages. In a 1000-node topology, D-Cliques require 98% less edges and 96% less total messages, with further possible gains using a small-world topology across cliques.

cs.LG↗

arXiv2020

FLeet: Online Federated Learning via Staleness Awareness and Performance Prediction

Federated Learning (FL) is very appealing for its privacy benefits: essentially, a global model is trained with updates computed on mobile devices while keeping the data of users local. Standard FL infrastructures are however designed to have no energy or performance impact on mobile devices, and are therefore not suitable for applications that require frequent (online) model updates, such as news recommenders. This paper presents FLeet, the first Online FL system, acting as a middleware between the Android OS and the machine learning application. FLeet combines the privacy of Standard FL with the precision of online learning thanks to two core components: (i) I-Prof, a new lightweight profiler that predicts and controls the impact of learning tasks on mobile devices, and (ii) AdaSGD, a new adaptive learning algorithm that is resilient to delayed updates. Our extensive evaluation shows that Online FL, as implemented by FLeet, can deliver a 2.3x quality boost compared to Standard FL, while only consuming 0.036% of the battery per day. I-Prof can accurately control the impact of learning tasks by improving the prediction accuracy up to 3.6x (computation time) and up to 19x (energy). AdaSGD outperforms alternative FL approaches by 18.4% in terms of convergence speed on heterogeneous data.

cs.LG↗

arXiv2020

Cluster-and-Conquer: When Randomness Meets Graph Locality

K-Nearest-Neighbors (KNN) graphs are central to many emblematic data mining and machine-learning applications. Some of the most efficient KNN graph algorithms are incremental and local: they start from a random graph, which they incrementally improve by traversing neighbors-of-neighbors links. Paradoxically, this random start is also one of the key weaknesses of these algorithms: nodes are initially connected to dissimilar neighbors, that lie far away according to the similarity metric. As a result, incremental algorithms must first laboriously explore spurious potential neighbors before they can identify similar nodes, and start converging. In this paper, we remove this drawback with Cluster-and-Conquer (C 2 for short). Cluster-and-Conquer boosts the starting configuration of greedy algorithms thanks to a novel lightweight clustering mechanism, dubbed FastRandomHash. FastRandomHash leverages random-ness and recursion to pre-cluster similar nodes at a very low cost. Our extensive evaluation on real datasets shows that Cluster-and-Conquer significantly outperforms existing approaches, including LSH, yielding speed-ups of up to x4.42 while incurring only a negligible loss in terms of KNN quality.

cs.DB↗

arXiv2019

Quicker ADC : Unlocking the hidden potential of Product Quantization with SIMD

Efficient Nearest Neighbor (NN) search in high-dimensional spaces is a foundation of many multimedia retrieval systems. A common approach is to rely on Product Quantization, which allows the storage of large vector databases in memory and efficient distance computations. Yet, implementations of nearest neighbor search with Product Quantization have their performance limited by the many memory accesses they perform. Following this observation, André et al. proposed Quick ADC with up to $6\times$ faster implementations of $m\times{}4$ product quantizers (PQ) leveraging specific SIMD instructions. Quicker ADC is a generalization of Quick ADC not limited to $m\times{}4$ codes and supporting AVX-512, the latest revision of SIMD instruction set. In doing so, Quicker ADC faces the challenge of using efficiently 5,6 and 7-bit shuffles that do not align to computer bytes or words. To this end, we introduce (i) irregular product quantizers combining sub-quantizers of different granularity and (ii) split tables allowing lookup tables larger than registers. We evaluate Quicker ADC with multiple indexes including Inverted Multi-Indexes and IVF HNSW and show that it outperforms the reference optimized implementations (i.e., FAISS and polysemous codes) for numerous configurations. Finally, we release an open-source fork of FAISS enhanced with Quicker ADC at http://github.com/nlescoua/faiss-quickeradc.

cs.CV↗

arXiv2019

Derived Codebooks for High-Accuracy Nearest Neighbor Search

High-dimensional Nearest Neighbor (NN) search is central in multimedia search systems. Product Quantization (PQ) is a widespread NN search technique which has a high performance and good scalability. PQ compresses high-dimensional vectors into compact codes thanks to a combination of quantizers. Large databases can, therefore, be stored entirely in RAM, enabling fast responses to NN queries. In almost all cases, PQ uses 8-bit quantizers as they offer low response times. In this paper, we advocate the use of 16-bit quantizers. Compared to 8-bit quantizers, 16-bit quantizers boost accuracy but they increase response time by a factor of 3 to 10. We propose a novel approach that allows 16-bit quantizers to offer the same response time as 8-bit quantizers, while still providing a boost of accuracy. Our approach builds on two key ideas: (i) the construction of derived codebooks that allow a fast and approximate distance evaluation, and (ii) a two-pass NN search procedure which builds a candidate set using the derived codebooks, and then refines it using 16-bit quantizers. On 1 billion SIFT vectors, with an inverted index, our approach offers a Recall@100 of 0.85 in 5.2 ms. By contrast, 16-bit quantizers alone offer a Recall@100 of 0.85 in 39 ms, and 8-bit quantizers a Recall@100 of 0.82 in 3.8 ms.

cs.CV↗

arXiv2018

Limiting the Spread of Fake News on Social Media Platforms by Evaluating Users' Trustworthiness

Today's social media platforms enable to spread both authentic and fake news very quickly. Some approaches have been proposed to automatically detect such "fake" news based on their content, but it is difficult to agree on universal criteria of authenticity (which can be bypassed by adversaries once known). Besides, it is obviously impossible to have each news item checked by a human. In this paper, we a mechanism to limit the spread of fake news which is not based on content. It can be implemented as a plugin on a social media platform. The principle is as follows: a team of fact-checkers reviews a small number of news items (the most popular ones), which enables to have an estimation of each user's inclination to share fake news items. Then, using a Bayesian approach, we estimate the trustworthiness of future news items, and treat accordingly those of them that pass a certain "untrustworthiness" threshold. We then evaluate the effectiveness and overhead of this technique on a large Twitter graph. We show that having a few thousands users exposed to one given news item enables to reach a very precise estimation of its reliability. We thus identify more than 99% of fake news items with no false positives. The performance impact is very small: the induced overhead on the 90th percentile latency is less than 3%, and less than 8% on the throughput of user operations.

cs.SI↗

arXiv2017

Accelerated Nearest Neighbor Search with Quick ADC

Efficient Nearest Neighbor (NN) search in high-dimensional spaces is a foundation of many multimedia retrieval systems. Because it offers low responses times, Product Quantization (PQ) is a popular solution. PQ compresses high-dimensional vectors into short codes using several sub-quantizers, which enables in-RAM storage of large databases. This allows fast answers to NN queries, without accessing the SSD or HDD. The key feature of PQ is that it can compute distances between short codes and high-dimensional vectors using cache-resident lookup tables. The efficiency of this technique, named Asymmetric Distance Computation (ADC), remains limited because it performs many cache accesses. In this paper, we introduce Quick ADC, a novel technique that achieves a 3 to 6 times speedup over ADC by exploiting Single Instruction Multiple Data (SIMD) units available in current CPUs. Efficiently exploiting SIMD requires algorithmic changes to the ADC procedure. Namely, Quick ADC relies on two key modifications of ADC: (i) the use 4-bit sub-quantizers instead of the standard 8-bit sub-quantizers and (ii) the quantization of floating-point distances. This allows Quick ADC to exceed the performance of state-of-the-art systems, e.g., it achieves a Recall@100 of 0.94 in 3.4 ms on 1 billion SIFT descriptors (128-bit codes).

cs.CV↗

arXiv2016

Bounds on the Voter Model in Dynamic Networks

In the voter model, each node of a graph has an opinion, and in every round each node chooses independently a random neighbour and adopts its opinion. We are interested in the consensus time, which is the first point in time where all nodes have the same opinion. We consider dynamic graphs in which the edges are rewired in every round (by an adversary) giving rise to the graph sequence $G_1, G_2, \dots $, where we assume that $G_i$ has conductance at least $ϕ_i$. We assume that the degrees of nodes don't change over time as one can show that the consensus time can become super-exponential otherwise. In the case of a sequence of $d$-regular graphs, we obtain asymptotically tight results. Even for some static graphs, such as the cycle, our results improve the state of the art. Here we show that the expected number of rounds until all nodes have the same opinion is bounded by $O(m/(d_{min} \cdot ϕ))$, for any graph with $m$ edges, conductance $ϕ$, and degrees at least $d_{min}$. In addition, we consider a biased dynamic voter model, where each opinion $i$ is associated with a probability $P_i$, and when a node chooses a neighbour with that opinion, it adopts opinion $i$ with probability $P_i$ (otherwise the node keeps its current opinion). We show for any regular dynamic graph, that if there is an $ε>0$ difference between the highest and second highest opinion probabilities, and at least $Ω(\log n)$ nodes have initially the opinion with the highest probability, then all nodes adopt w.h.p. that opinion. We obtain a bound on the convergences time, which becomes $O(\log n/ϕ)$ for static graphs.

cs.SI↗

arXiv2015

Beyond One Third Byzantine Failures

The Byzantine agreement problem requires a set of $n$ processes to agree on a value sent by a transmitter, despite a subset of $b$ processes behaving in an arbitrary, i.e. Byzantine, manner and sending corrupted messages to all processes in the system. It is well known that the problem has a solution in a (an eventually) synchronous message passing distributed system iff the number of processes in the Byzantine subset is less than one third of the total number of processes, i.e. iff $n > 3b+1$. The rest of the processes are expected to be correct: they should never deviate from the algorithm assigned to them and send corrupted messages. But what if they still do? We show in this paper that it is possible to solve Byzantine agreement even if, beyond the $ b$ ($< n/3 $) Byzantine processes, some of the other processes also send corrupted messages, as long as they do not send them to all. More specifically, we generalize the classical Byzantine model and consider that Byzantine failures might be partial. In each communication step, some of the processes might send corrupted messages to a subset of the processes. This subset of processes - to which corrupted messages might be sent - could change over time. We compute the exact number of processes that can commit such faults, besides those that commit classical Byzantine failures, while still solving Byzantine agreement. We present a corresponding Byzantine agreement algorithm and prove its optimality by giving resilience and complexity bounds.

cs.DC↗

arXiv2015

Heterogeneous Differential Privacy

The massive collection of personal data by personalization systems has rendered the preservation of privacy of individuals more and more difficult. Most of the proposed approaches to preserve privacy in personalization systems usually address this issue uniformly across users, thus ignoring the fact that users have different privacy attitudes and expectations (even among their own personal data). In this paper, we propose to account for this non-uniformity of privacy expectations by introducing the concept of heterogeneous differential privacy. This notion captures both the variation of privacy expectations among users as well as across different pieces of information related to the same user. We also describe an explicit mechanism achieving heterogeneous differential privacy, which is a modification of the Laplacian mechanism by Dwork, McSherry, Nissim, and Smith. In a nutshell, this mechanism achieves heterogeneous differential privacy by manipulating the sensitivity of the function using a linear transformation on the input domain. Finally, we evaluate on real datasets the impact of the proposed mechanism with respect to a semantic clustering task. The results of our experiments demonstrate that heterogeneous differential privacy can account for different privacy attitudes while sustaining a good level of utility as measured by the recall for the semantic clustering task.

cs.CR↗

arXiv2014

Signed graph embedding: when everybody can sit closer to friends than enemies

Signed graphs are graphs with signed edges. They are commonly used to represent positive and negative relationships in social networks. While balance theory and clusterizable graphs deal with signed graphs to represent social interactions, recent empirical studies have proved that they fail to reflect some current practices in real social networks. In this paper we address the issue of drawing signed graphs and capturing such social interactions. We relax the previous assumptions to define a drawing as a model in which every vertex has to be placed closer to its neighbors connected via a positive edge than its neighbors connected via a negative edge in the resulting space. Based on this definition, we address the problem of deciding whether a given signed graph has a drawing in a given $\ell$-dimensional Euclidean space. We present forbidden patterns for signed graphs that admit the introduced definition of drawing in the Euclidean plane and line. We then focus on the $1$-dimensional case, where we provide a polynomial time algorithm that decides if a given complete signed graph has a drawing, and constructs it when applicable.

cs.DM↗

arXiv2013

Repairing Multiple Failures with Coordinated and Adaptive Regenerating Codes

Erasure correcting codes are widely used to ensure data persistence in distributed storage systems. This paper addresses the simultaneous repair of multiple failures in such codes. We go beyond existing work (i.e., regenerating codes by Dimakis et al.) by describing (i) coordinated regenerating codes (also known as cooperative regenerating codes) which support the simultaneous repair of multiple devices, and (ii) adaptive regenerating codes which allow adapting the parameters at each repair. Similarly to regenerating codes by Dimakis et al., these codes achieve the optimal tradeoff between storage and the repair bandwidth. Based on these extended regenerating codes, we study the impact of lazy repairs applied to regenerating codes and conclude that lazy repairs cannot reduce the costs in term of network bandwidth but allow reducing the disk-related costs (disk bandwidth and disk I/O).

cs.IT↗

arXiv2013

Regenerating Codes: A System Perspective

The explosion of the amount of data stored in cloud systems calls for more efficient paradigms for redundancy. While replication is widely used to ensure data availability, erasure correcting codes provide a much better trade-off between storage and availability. Regenerating codes are good candidates for they also offer low repair costs in term of network bandwidth. While they have been proven optimal, they are difficult to understand and parameterize. In this paper we provide an analysis of regenerating codes for practitioners to grasp the various trade-offs. More specifically we make two contributions: (i) we study the impact of the parameters by conducting an analysis at the level of the system, rather than at the level of a single device; (ii) we compare the computational costs of various implementations of codes and highlight the most efficient ones. Our goal is to provide system designers with concrete information to help them choose the best parameters and design for regenerating codes.

cs.DC↗

arXiv2013

On Dynamic Distributed Computing

This paper shows for the first time that distributed computing can be both reliable and efficient in an environment that is both highly dynamic and hostile. More specifically, we show how to maintain clusters of size $O(\log N)$, each containing more than two thirds of honest nodes with high probability, within a system whose size can vary \textit{polynomially} with respect to its initial size. Furthermore, the communication cost induced by each node arrival or departure is polylogarithmic with respect to $N$, the maximal size of the system. Our clustering can be achieved despite the presence of a Byzantine adversary controlling a fraction $\bad \leq \{1}{3}-ε$ of the nodes, for some fixed constant $ε> 0$, independent of $N$. So far, such a clustering could only be performed for systems who size can vary constantly and it was not clear whether that was at all possible for polynomial variances.

cs.DC↗