Searcharxiv⌕ Search

arXiv subjects

Pablo Piantanida

Publications and source records attributed to Pablo Piantanida.

At least 91 records · Page 5Linked to original sources
arXiv2021

Combination Networks with End-user-caches: Novel Achievable and Converse Bounds under Uncoded Cache Placement

Caching is an efficient way to reduce network traffic congestion during peak hours by storing some content at the users' local caches. For the shared-link network with end-user-caches, Maddah-Ali and Niesen proposed a two-phase coded caching strategy. In practice, users may communicate with the server through intermediate relays. This paper studies the tradeoff between the memory size $M$ and the network load $R$ for networks where a server with $N$ files is connected to $H$ relays (without caches), which in turn are connected to $K$ users equipped with caches of $M$ files. When each user is connected to a different subset of $r$ relays, i.e., $K = \binom{H}{r}$, the system is referred to as a {\it combination network with end-user-caches}. In this work, converse bounds are derived for the practically motivated case of {\it uncoded} cache contents, that is, bits of the various files are directly pushed into the user caches without any coding. In this case, once the cache contents and the user demands are known, the problem reduces to a general index coding problem.This paper shows that relying on a well-known "acyclic index coding converse bound" results in converse bounds that are not tight for combination networks with end-user-caches. A novel converse bound that leverages the network topology is proposed, which is the tightest converse bound known to date. As a result of independent interest, an inequality that generalizes the well-known sub-modularity of entropy is derived. Several novel caching schemes are proposed, based on the Maddah-Ali and Niesen cache placement. The proposed schemes are proved: (i) to be (order) optimal for some $(N,M,H,r)$ parameters regimes under the constraint of uncoded cache placement, and (ii) to outperform the state-of-the-art schemes in numerical evaluations.

cs.IT↗
arXiv2021

DOCTOR: A Simple Method for Detecting Misclassification Errors

Deep neural networks (DNNs) have shown to perform very well on large scale object recognition problems and lead to widespread use for real-world applications, including situations where DNN are implemented as "black boxes". A promising approach to secure their use is to accept decisions that are likely to be correct while discarding the others. In this work, we propose DOCTOR, a simple method that aims to identify whether the prediction of a DNN classifier should (or should not) be trusted so that, consequently, it would be possible to accept it or to reject it. Two scenarios are investigated: Totally Black Box (TBB) where only the soft-predictions are available and Partially Black Box (PBB) where gradient-propagation to perform input pre-processing is allowed. Empirically, we show that DOCTOR outperforms all state-of-the-art methods on various well-known images and sentiment analysis datasets. In particular, we observe a reduction of up to $4\%$ of the false rejection rate (FRR) in the PBB scenario. DOCTOR can be applied to any pre-trained model, it does not require prior information about the underlying dataset and is as simple as the simplest available methods in the literature.

cs.CV↗
arXiv2021

Automatic Text Evaluation through the Lens of Wasserstein Barycenters

A new metric \texttt{BaryScore} to evaluate text generation based on deep contextualized embeddings e.g., BERT, Roberta, ELMo) is introduced. This metric is motivated by a new framework relying on optimal transport tools, i.e., Wasserstein distance and barycenter. By modelling the layer output of deep contextualized embeddings as a probability distribution rather than by a vector embedding; this framework provides a natural way to aggregate the different outputs through the Wasserstein space topology. In addition, it provides theoretical grounds to our metric and offers an alternative to available solutions e.g., MoverScore and BertScore). Numerical evaluation is performed on four different tasks: machine translation, summarization, data2text generation and image captioning. Our results show that \texttt{BaryScore} outperforms other BERT based metrics and exhibits more consistent behaviour in particular for text summarization.

cs.CL↗
arXiv2021

Learning Sparse Privacy-Preserving Representations for Smart Meters Data

Fine-grained Smart Meters (SMs) data recording and communication has enabled several features of Smart Grids (SGs) such as power quality monitoring, load forecasting, fault detection, and so on. In addition, it has benefited the users by giving them more control over their electricity consumption. However, it is well-known that it also discloses sensitive information about the users, i.e., an attacker can infer users' private information by analyzing the SMs data. In this study, we propose a privacy-preserving approach based on non-uniform down-sampling of SMs data. We formulate this as the problem of learning a sparse representation of SMs data with minimum information leakage and maximum utility. The architecture is composed of a releaser, which is a recurrent neural network (RNN), that is trained to generate the sparse representation by masking the SMs data, and an utility and adversary networks (also RNNs), which help the releaser to minimize the leakage of information about the private attribute, while keeping the reconstruction error of the SMs data minimum (i.e., maximum utility). The performance of the proposed technique is assessed based on actual SMs data and compared with uniform down-sampling, random (non-uniform) down-sampling, as well as the state-of-the-art in privacy-preserving methods using a data manipulation approach. It is shown that our method performs better in terms of the privacy-utility trade-off while releasing much less data, thus also being more efficient.

eess.SP↗
arXiv2021

On Universal D-Semifaithful Coding for Memoryless Sources with Infinite Alphabets

The problem of variable length and fixed-distortion universal source coding (or D-semifaithful source coding) for stationary and memoryless sources on countably infinite alphabets ($\infty$-alphabets) is addressed in this paper. The main results of this work offer a set of sufficient conditions (from weaker to stronger) to obtain weak minimax universality, strong minimax universality, and corresponding achievable rates of convergences for the worse-case redundancy for the family of stationary memoryless sources whose densities are dominated by an envelope function (or the envelope family) on $\infty$-alphabets. An important implication of these results is that universal D-semifaithful source coding is not feasible for the complete family of stationary and memoryless sources on $\infty$-alphabets. To demonstrate this infeasibility, a sufficient condition for the impossibility is presented for the envelope family. Interestingly, it matches the well-known impossibility condition in the context of lossless (variable-length) universal source coding. More generally, this work offers a simple description of what is needed to achieve universal D-semifaithful coding for a family of distributions $Λ$. This reduces to finding a collection of quantizations of the product space at different block-lengths -- reflecting the fixed distortion restriction -- that satisfy two asymptotic requirements: the first is a universal quantization condition with respect to $Λ$, and the second is a vanishing information radius (I-radius) condition for $Λ$ reminiscent of the condition known for lossless universal source coding.

cs.IT↗
arXiv2021

Mutual-Information Based Few-Shot Classification

We introduce Transductive Infomation Maximization (TIM) for few-shot learning. Our method maximizes the mutual information between the query features and their label predictions for a given few-shot task, in conjunction with a supervision loss based on the support set. We motivate our transductive loss by deriving a formal relation between the classification accuracy and mutual-information maximization. Furthermore, we propose a new alternating-direction solver, which substantially speeds up transductive inference over gradient-based optimization, while yielding competitive accuracy. We also provide a convergence analysis of our solver based on Zangwill's theory and bound-optimization arguments. TIM inference is modular: it can be used on top of any base-training feature extractor. Following standard transductive few-shot settings, our comprehensive experiments demonstrate that TIM outperforms state-of-the-art methods significantly across various datasets and networks, while used on top of a fixed feature extractor trained with simple cross-entropy on the base classes, without resorting to complex meta-learning schemes. It consistently brings between 2 % and 5 % improvement in accuracy over the best performing method, not only on all the well-established few-shot benchmarks but also on more challenging scenarios, with random tasks, domain shift and larger numbers of classes, as in the recently introduced META-DATASET. Our code is publicly available at https://github.com/mboudiaf/TIM. We also publicly release a standalone PyTorch implementation of META-DATASET, along with additional benchmarking results, at https://github.com/mboudiaf/pytorch-meta-dataset.

cs.CV↗
arXiv2021

A Novel Estimator of Mutual Information for Learning to Disentangle Textual Representations

Learning disentangled representations of textual data is essential for many natural language tasks such as fair classification, style transfer and sentence generation, among others. The existent dominant approaches in the context of text data {either rely} on training an adversary (discriminator) that aims at making attribute values difficult to be inferred from the latent code {or rely on minimising variational bounds of the mutual information between latent code and the value attribute}. {However, the available methods suffer of the impossibility to provide a fine-grained control of the degree (or force) of disentanglement.} {In contrast to} {adversarial methods}, which are remarkably simple, although the adversary seems to be performing perfectly well during the training phase, after it is completed a fair amount of information about the undesired attribute still remains. This paper introduces a novel variational upper bound to the mutual information between an attribute and the latent code of an encoder. Our bound aims at controlling the approximation error via the Renyi's divergence, leading to both better disentangled representations and in particular, a precise control of the desirable degree of disentanglement {than state-of-the-art methods proposed for textual data}. Furthermore, it does not suffer from the degeneracy of other losses in multi-class scenarios. We show the superiority of this method on fair classification and on textual style transfer tasks. Additionally, we provide new insights illustrating various trade-offs in style transfer when attempting to learn disentangled representations and quality of the generated sentence.

cs.AI↗
arXiv2021

Few-Shot Segmentation Without Meta-Learning: A Good Transductive Inference Is All You Need?

We show that the way inference is performed in few-shot segmentation tasks has a substantial effect on performances -- an aspect often overlooked in the literature in favor of the meta-learning paradigm. We introduce a transductive inference for a given query image, leveraging the statistics of its unlabeled pixels, by optimizing a new loss containing three complementary terms: i) the cross-entropy on the labeled support pixels; ii) the Shannon entropy of the posteriors on the unlabeled query-image pixels; and iii) a global KL-divergence regularizer based on the proportion of the predicted foreground. As our inference uses a simple linear classifier of the extracted features, its computational load is comparable to inductive inference and can be used on top of any base training. Foregoing episodic training and using only standard cross-entropy training on the base classes, our inference yields competitive performances on standard benchmarks in the 1-shot scenarios. As the number of available shots increases, the gap in performances widens: on PASCAL-5i, our method brings about 5% and 6% improvements over the state-of-the-art, in the 5- and 10-shot scenarios, respectively. Furthermore, we introduce a new setting that includes domain shifts, where the base and novel classes are drawn from different datasets. Our method achieves the best performances in this more realistic setting. Our code is freely available online: https://github.com/mboudiaf/RePRI-for-Few-Shot-Segmentation.

cs.CV↗
arXiv2020

Transductive Information Maximization For Few-Shot Learning

We introduce Transductive Infomation Maximization (TIM) for few-shot learning. Our method maximizes the mutual information between the query features and their label predictions for a given few-shot task, in conjunction with a supervision loss based on the support set. Furthermore, we propose a new alternating-direction solver for our mutual-information loss, which substantially speeds up transductive-inference convergence over gradient-based optimization, while yielding similar accuracy. TIM inference is modular: it can be used on top of any base-training feature extractor. Following standard transductive few-shot settings, our comprehensive experiments demonstrate that TIM outperforms state-of-the-art methods significantly across various datasets and networks, while used on top of a fixed feature extractor trained with simple cross-entropy on the base classes, without resorting to complex meta-learning schemes. It consistently brings between 2% and 5% improvement in accuracy over the best performing method, not only on all the well-established few-shot benchmarks but also on more challenging scenarios,with domain shifts and larger numbers of classes.

cs.LG↗
arXiv2020

Fundamental Limits of Decentralized Data Shuffling

Data shuffling of training data among different computing nodes (workers) has been identified as a core element to improve the statistical performance of modern large-scale machine learning algorithms. Data shuffling is often considered as one of the most significant bottlenecks in such systems due to the heavy communication load. Under a master-worker architecture (where a master has access to the entire dataset and only communication between the master and the workers is allowed) coding has been recently proved to considerably reduce the communication load. This work considers a different communication paradigm referred to as decentralized data shuffling, where workers are allowed to communicate with one another via a shared link. The decentralized data shuffling problem has two phases: workers communicate with each other during the data shuffling phase, and then workers update their stored content during the storage phase. The main challenge is to derive novel converse bounds and achievable schemes for decentralized data shuffling by considering the asymmetry of the workers' storages (i.e., workers are constrained to store different files in their storages based on the problem setting), in order to characterize the fundamental limits of this problem. For the case of uncoded storage (i.e., each worker directly stores a subset of bits of the dataset), this paper proposes converse and achievable bounds (based on distributed interference alignment and distributed clique-covering strategies) that are within a factor of 3/2 of one another. The proposed schemes are also exactly optimal under the constraint of uncoded storage for either large storage size or at most four workers in the system.

cs.IT↗
arXiv2020

An Index Coding Approach to Caching with Uncoded Cache Placement

Caching is an efficient way to reduce network traffic congestion during peak hours, by storing some content at the user's local cache memory, even without knowledge of user's later demands. Maddah-Ali and Niesen proposed a two-phase (placement phase and delivery phase) coded caching strategy for broadcast channels with cache-aided users. This paper investigates the same model under the constraint that content is placed uncoded within the caches, that is, when bits of the files are simply copied within the caches. When the cache contents are uncoded and the users' demands are revealed, the caching problem can be connected to an index coding problem. This paper focuses on deriving fundamental performance limits for the caching problem by using tools for the index coding problem that were either known or are newly developed in this work. First, a converse bound for the caching problem under the constraint of uncoded cache placement is proposed based on the "acyclic index coding converse bound". This converse bound is proved to be achievable by the Maddah-Ali and Niesen's scheme when the number of files is not less than the number of users, and by a newly derived index coding achievable scheme otherwise. The proposed index coding achievable scheme is based on distributed source coding and strictly improves on the widely used "composite (index) coding" achievable bound and its improvements, and is of independent interest. An important consequence of the findings of this paper is that advancements on the coded caching problem posed by Maddah-Ali and Niesen are thus only possible by considering strategies with coded placement phase. A recent work by Yu et al. has however shown that coded cache placement can at most half the network load compared to the results presented in this paper.

cs.IT↗
arXiv2019

Key and Message Semantic-Security over State-Dependent Channels

We study the trade-off between secret message (SM) and secret key (SK) rates, simultaneously achievable over a state-dependent (SD) wiretap channel (WTC) with non-causal channel state information (CSI) at the encoder. This model subsumes other instances of CSI availability as special cases, and calls for efficient utilization of the state sequence for both reliability and security purposes. An inner bound on the semantic-security (SS) SM-SK capacity region is derived based on a superposition coding scheme inspired by a past work of the authors. The region is shown to attain capacity for a certain class of SD-WTCs. SS is established by virtue of two versions of the strong soft-covering lemma. The derived region yields an improvement upon the previously best known SM-SK trade-off result reported by Prabhakaran et al., and, to the best of our knowledge, upon all other existing lower bounds for either SM or SK for this setup, even if the semantic security requirement is relaxed to weak secrecy. It is demonstrated that our region can be strictly larger than those reported in the preceding works.

cs.IT↗
arXiv2019

Understanding the Behaviour of the Empirical Cross-Entropy Beyond the Training Distribution

Machine learning theory has mostly focused on generalization to samples from the same distribution as the training data. Whereas a better understanding of generalization beyond the training distribution where the observed distribution changes is also fundamentally important to achieve a more powerful form of generalization. In this paper, we attempt to study through the lens of information measures how a particular architecture behaves when the true probability law of the samples is potentially different at training and testing times. Our main result is that the testing gap between the empirical cross-entropy and its statistical expectation (measured with respect to the testing probability law) can be bounded with high probability by the mutual information between the input testing samples and the corresponding representations, generated by the encoder obtained at training time. These results of theoretical nature are supported by numerical simulations showing that the mentioned mutual information is representative of the testing gap, capturing qualitatively the dynamic in terms of the hyperparameters of the network.

stat.ML↗
arXiv2018

Secret Key Generation over Noisy Channels with Correlated Sources

This paper investigates the problem of secret key generation over a wiretap channel when the terminals observe correlated sources. These sources are independent of the main channel and the users overhear them before the transmission takes place. A novel achievable scheme is proposed, and its optimality is shown under certain less-noisy conditions. This result improves upon the existing literature where the more stringent condition of degradedness is required. Furthermore, numerical evaluation of the proposed scheme and previously reported results for a binary model are presented; a comparison of the numerical bounds provides insights on the benefit of the novel scheme.

cs.IT↗
arXiv2018

The Wiretap Channel with Generalized Feedback: Secure Communication and Key Generation

It is a well-known fact that feedback does not increase the capacity of point-to-point memoryless channels, however, its effect in secure communications is not fully understood yet. In this work, an achievable scheme for the wiretap channel with generalized feedback is presented. This scheme, which uses the feedback signal to generate a shared secret key between the legitimate users, encrypts the message to be sent at the bit level. New capacity results for a class of channels are provided, as well as some new insights into the secret key agreement problem. Moreover, this scheme recovers previously reported rate regions from the literature, and thus it can be seen as a generalization that unifies several results in the field.

cs.IT↗
arXiv2018

Caching in Combination Networks: Novel Multicast Message Generation and Delivery by Leveraging the Network Topology

Maddah-Ali and Niesen's original coded caching scheme for shared-link broadcast networks is now known to be optimal to within a factor two, and has been applied to other types of networks. For practical reasons, this paper considers that a server communicates to cache-aided users through $H$ intermediate relays. In particular, it focuses on combination networks where each of the $K = \binom{H}{r}$ users is connected to a distinct $r$-subsets of relays. By leveraging the symmetric topology of the network, this paper proposes a novel method to general multicast messages and to deliver them to the users. By numerical evaluations, the proposed scheme is shown to reduce the download time compared to the schemes available in the literature. The idea is then extended to decentralized combination networks, more general relay networks, and combination networks with cache-aided relays and users. Also in these cases the proposed scheme outperforms known ones.

cs.IT↗
arXiv2018

On Combination Networks with Cache-aided Relays and Users

Caching is an efficient way to reduce peak hour network traffic congestion by storing some contents at the user's cache without knowledge of later demands. Coded caching strategy was originally proposed by Maddah-Ali and Niesen to give an additional coded caching gain compared the conventional uncoded scheme. Under practical consideration, the caching model was recently considered in relay network, in particular the combination network, where the central server communicates with $K=\binom{H}{r}$ users (each is with a cache of $M$ files) through $H$ immediate relays, and each user is connected to a different $r-$subsets of relays. Several inner bounds and outer bounds were proposed for combination networks with end-user-caches. This paper extends the recent work by the authors on centralized combination networks with end-user caches to a more general setting, where both relays and users have caches. In contrast to the existing schemes in which the packets transmitted from the server are independent of the cached contents of relays, we propose a novel caching scheme by creating an additional coded caching gain to the transmitted load from the server with the help of the cached contents in relays. We also show that the proposed scheme outperforms the state-of-the-art approaches.

cs.IT↗
arXiv2018

On the Benefits of Asymmetric Coded Cache Placement in Combination Networks with End-User Caches

This paper investigates the fundamental tradeoff between cache size and download time in the (H;r;M;N) combination network, where a server with N files is connected to H relays (without caches) and each of the K:=\binom{H}{r} users (with caches of size M files) is connected to a different subset of r relays. Existing schemes fall within two categories: either use the uncoded symmetric cache placement originally proposed for the shared-link model and design delivery phase dependent on the network topology, or effectively divide the combination network into H independent shared-link networks each serving \binom{H-1}{r-1} users; in either case, the placement phase is independent of network topology. In this paper, a novel strategy is proposed where the coded cache placement is dependent on network topology. The proposed scheme is shown to be information theoretically optimal for large cache size. In addition, when not exactly optimal, the proposed scheme can also outperform existing schemes.

cs.IT↗