arXiv subjects

Roger Wattenhofer

Publications and source records attributed to Roger Wattenhofer.

At least 235 records · Page 13Linked to original sources

arXiv2020

Brick: Asynchronous Payment Channels

Off-chain protocols (channels) are a promising solution to the scalability and privacy challenges of blockchain payments. Current proposals, however, require synchrony assumptions to preserve the safety of a channel, leaking to an adversary the exact amount of time needed to control the network for a successful attack. In this paper, we introduce Brick, the first payment channel that remains secure under network asynchrony and concurrently provides correct incentives. The core idea is to incorporate the conflict resolution process within the channel by introducing a rational committee of external parties, called Wardens. Hence, if a party wants to close a channel unilaterally, it can only get the committee's approval for the last valid state. Brick provides sub-second latency because it does not employ heavy-weight consensus. Instead, Brick uses consistent broadcast to announce updates and close the channel, a light-weight abstraction that is powerful enough to preserve safety and liveness to any rational parties. Furthermore, we consider permissioned blockchains, where the additional property of auditability might be desired for regulatory purposes. We introduce Brick+, an off-chain construction that provides auditability on top of Brick without conflicting with its privacy guarantees. We formally define the properties our payment channel construction should fulfill, and prove that both Brick and Brick+ satisfy them. We also design incentives for Brick such that honest and rational behavior aligns. Finally, we provide a reference implementation of the smart contracts in Solidity.

cs.DC↗

arXiv2020

Asynchronous Byzantine Agreement in Incomplete Networks [Technical Report]

The Byzantine agreement problem is considered to be a core problem in distributed systems. For example, Byzantine agreement is needed to build a blockchain, a totally ordered log of records. Blockchains are asynchronous distributed systems, fault-tolerant against Byzantine nodes. In the literature, the asynchronous byzantine agreement problem is studied in a fully connected network model where every node can directly send messages to every other node. This assumption is questionable in many real-world environments. In the reality, nodes might need to communicate by means of an incomplete network, and Byzantine nodes might not forward messages. Furthermore, Byzantine nodes might not behave correctly and, for example, corrupt messages. Therefore, in order to truly understand Byzantine Agreement, we need both ingredients: asynchrony and incomplete communication networks. In this paper, we study the asynchronous Byzantine agreement problem in incomplete networks. A classic result by Danny Dolev proved that in a distributed system with n nodes in the presence of f Byzantine nodes, the vertex connectivity of the system communication graph should be at least (2f+1). While Dolev's result was for synchronous deterministic systems, we demonstrate that the same bound also holds for asynchronous randomized systems. We show that the bound is tight by presenting a randomized algorithm, and a matching lower bound.

cs.DC↗

arXiv2020

Normalized Attention Without Probability Cage

Attention architectures are widely used; they recently gained renewed popularity with Transformers yielding a streak of state of the art results. Yet, the geometrical implications of softmax-attention remain largely unexplored. In this work we highlight the limitations of constraining attention weights to the probability simplex and the resulting convex hull of value vectors. We show that Transformers are sequence length dependent biased towards token isolation at initialization and contrast Transformers to simple max- and sum-pooling - two strong baselines rarely reported. We propose to replace the softmax in self-attention with normalization, yielding a hyperparameter and data-bias robust, generally applicable architecture. We support our insights with empirical results from more than 25,000 trained models. All results and implementations are made available.

cs.LG↗

arXiv2020

On the Hardness of Red-Blue Pebble Games

Red-blue pebble games model the computation cost of a two-level memory hierarchy. We present various hardness results in different red-blue pebbling variants, with a focus on the oneshot model. We first study the relationship between previously introduced red-blue pebble models (base, oneshot, nodel). We also analyze a new variant (compcost) to obtain a more realistic model of computation. We then prove that red-blue pebbling is NP-hard in all of these model variants. Furthermore, we show that in the oneshot model, a $δ$-approximation algorithm for $δ<2$ is only possible if the unique games conjecture is false. Finally, we show that greedy algorithms are not good candidates for approximation, since they can return significantly worse solutions than the optimum.

cs.CC↗

arXiv2020

A General Stabilization Bound for Influence Propagation in Graphs

We study the stabilization time of a wide class of processes on graphs, in which each node can only switch its state if it is motivated to do so by at least a $\frac{1+λ}{2}$ fraction of its neighbors, for some $0 < λ< 1$. Two examples of such processes are well-studied dynamically changing colorings in graphs: in majority processes, nodes switch to the most frequent color in their neighborhood, while in minority processes, nodes switch to the least frequent color in their neighborhood. We describe a non-elementary function $f(λ)$, and we show that in the sequential model, the worst-case stabilization time of these processes can completely be characterized by $f(λ)$. More precisely, we prove that for any $ε>0$, $O(n^{1+f(λ)+ε})$ is an upper bound on the stabilization time of any proportional majority/minority process, and we also show that there are graph constructions where stabilization indeed takes $Ω(n^{1+f(λ)-ε})$ steps.

cs.DM↗

arXiv2020

Online Graph Exploration on a Restricted Graph Class: Optimal Solutions for Tadpole Graphs

We study the problem of online graph exploration on undirected graphs, where a searcher has to visit every vertex and return to the origin. Once a new vertex is visited, the searcher learns of all neighboring vertices and the connecting edge weights. The goal such an exploration is to minimize its total cost, where each edge traversal incurs a cost of the corresponding edge weight. We investigate the problem on tadpole graphs (also known as dragons, kites), which consist of a cycle with an attached path. Miyazaki et al. (The online graph exploration problem on restricted graphs, IEICE Transactions 92-D (9), 2009) showed that every online algorithm on these graphs must have a competitive ratio of 2-epsilon, but did not provide upper bounds for non-unit edge weights. We show via amortized analysis that a greedy approach yields a matching competitive ratio of 2 on tadpole graphs, for arbitrary non-negative edge weights.

cs.DS↗

arXiv2020

Network-Aware Strategies in Financial Systems

We study the incentives of banks in a financial network, where the network consists of debt contracts and credit default swaps (CDSs) between banks. One of the most important questions in such a system is the problem of deciding which of the banks are in default, and how much of their liabilities these banks can pay. We study the payoff and preferences of the banks in the different solutions to this problem. We also introduce a more refined model which allows assigning priorities to payment obligations; this provides a more expressive and realistic model of real-life financial systems, while it always ensures the existence of a solution. The main focus of the paper is an analysis of the actions that a single bank can execute in a financial system in order to influence the outcome to its advantage. We show that removing an incoming debt, or donating funds to another bank can result in a single new solution that is strictly more favorable to the acting bank. We also show that increasing the bank's external funds or modifying the priorities of outgoing payments cannot introduce a more favorable new solution into the system, but may allow the bank to remove some unfavorable solutions, or to increase its recovery rate. Finally, we show how the actions of two banks in a simple financial system can result in classical game theoretic situations like the prisoner's dilemma or the dollar auction, demonstrating the wide expressive capability of the financial system model.

q-fin.RM↗

arXiv2020

On Identifiability in Transformers

In this paper we delve deep in the Transformer architecture by investigating two of its core components: self-attention and contextual embeddings. In particular, we study the identifiability of attention weights and token embeddings, and the aggregation of context into hidden tokens. We show that, for sequences longer than the attention head dimension, attention weights are not identifiable. We propose effective attention as a complementary tool for improving explanatory interpretations based on attention. Furthermore, we show that input tokens retain to a large degree their identity across the model. We also find evidence suggesting that identity information is mainly encoded in the angle of the embeddings and gradually decreases with depth. Finally, we demonstrate strong mixing of input information in the generation of contextual embeddings by means of a novel quantification method based on gradient attribution. Overall, we show that self-attention distributions are not directly interpretable and present tools to better understand and further investigate Transformer models.

cs.CL↗

arXiv2019

Ride the Lightning: The Game Theory of Payment Channels

Payment channels were introduced to solve various eminent cryptocurrency scalability issues. Multiple payment channels build a network on top of a blockchain, the so-called layer 2. In this work, we analyze payment networks through the lens of network creation games. We identify betweenness and closeness centrality as central concepts regarding payment networks. We study the topologies that emerge when players act selfishly and determine the parameter space in which they constitute a Nash equilibrium. Moreover, we determine the social optima depending on the correlation of betweenness and closeness centrality. When possible, we bound the price of anarchy. We also briefly discuss the price of stability.

cs.GT↗

arXiv2019

Synthetic Epileptic Brain Activities Using Generative Adversarial Networks

Epilepsy is a chronic neurological disorder affecting more than 65 million people worldwide and manifested by recurrent unprovoked seizures. The unpredictability of seizures not only degrades the quality of life of the patients, but it can also be life-threatening. Modern systems monitoring electroencephalography (EEG) signals are being currently developed with the view to detect epileptic seizures in order to alert caregivers and reduce the impact of seizures on patients' quality of life. Such seizure detection systems employ state-of-the-art machine learning algorithms that require a considerably large amount of labeled personal data for training. However, acquiring EEG signals of epileptic seizures is a costly and time-consuming process for medical experts and patients, currently requiring in-hospital recordings in specialized units. In this work, we generate synthetic seizure-like brain electrical activities, i.e., EEG signals, that can be used to train seizure detection algorithms, alleviating the need for recorded data. First, we train a Generative Adversarial Network (GAN) with data from 30 epilepsy patients. Then, we generate synthetic personalized training sets for new, unseen patients, which overall yield higher detection performance than the real-data training sets. We demonstrate our results using the datasets from the EPILEPSIAE Project, one of the world's largest public databases for seizure detection.

eess.SP↗

arXiv2019

Learning Policies through Quantile Regression

Policy gradient based reinforcement learning algorithms coupled with neural networks have shown success in learning complex policies in the model free continuous action space control setting. However, explicitly parameterized policies are limited by the scope of the chosen parametric probability distribution. We show that alternatively to the likelihood based policy gradient, a related objective can be optimized through advantage weighted quantile regression. Our approach models the policy implicitly in the network, which gives the agent the freedom to approximate any distribution in each action dimension, not limiting its capabilities to the commonly used unimodal Gaussian parameterization. This broader spectrum of policies makes our algorithm suitable for problems where Gaussian policies cannot fit the optimal policy. Moreover, our results on the MuJoCo physics simulator benchmarks are comparable or superior to state-of-the-art on-policy methods.

cs.LG↗

arXiv2019

Payment Networks as Creation Games

Payment networks were introduced to address the limitation on the transaction throughput of popular blockchains. To open a payment channel one has to publish a transaction on-chain and pay the appropriate transaction fee. A transaction can be routed in the network, as long as there is a path of channels with the necessary capital. The intermediate nodes on this path can ask for a fee to forward the transaction. Hence, opening channels, although costly, can benefit a party, both by reducing the cost of the party for sending a transaction and by collecting the fees from forwarding transactions of other parties. This trade-off spawns a network creation game between the channel parties. In this work, we introduce the first game theoretic model for analyzing the network creation game on blockchain payment channels. Further, we examine various network structures (path, star, complete bipartite graph and clique) and determine for each one of them the constraints (fee value) under which they constitute a Nash equilibrium, given a fixed fee policy. Last, we show that the star is a Nash equilibrium when each channel party can freely decide the channel fee. On the other hand, we prove the complete bipartite graph can never be a Nash equilibrium, given a free fee policy.

cs.GT↗

arXiv2019

Bitcoin Security under Temporary Dishonest Majority

We prove Bitcoin is secure under temporary dishonest majority. We assume the adversary can corrupt a specific fraction of parties and also introduce crash failures, i.e., some honest participants are offline during the execution of the protocol. We demand a majority of honest online participants on expectation. We explore three different models and present the requirements for proving Bitcoin's security in all of them: we first examine a synchronous model, then extend to a bounded delay model and last we consider a synchronous model that allows message losses.

cs.CR↗

arXiv2019

Online Payment Network Design

Payment channels allow transactions between participants of the blockchain to be executed securely off-chain, and thus provide a promising solution for the scalability problem of popular blockchains. We study the online network design problem for payment channels, assuming a central coordinator. We focus on a single channel, where the coordinator desires to maximize the number of accepted transactions under given capital constraints. Despite the simplicity of the problem, we present a flurry of impossibility results, both for deterministic and randomized algorithms against adaptive as well as oblivious adversaries.

cs.DS↗

arXiv2019

Attentive Multi-Task Deep Reinforcement Learning

Sharing knowledge between tasks is vital for efficient learning in a multi-task setting. However, most research so far has focused on the easier case where knowledge transfer is not harmful, i.e., where knowledge from one task cannot negatively impact the performance on another task. In contrast, we present an approach to multi-task deep reinforcement learning based on attention that does not require any a-priori assumptions about the relationships between tasks. Our attention network automatically groups task knowledge into sub-networks on a state level granularity. It thereby achieves positive knowledge transfer if possible, and avoids negative transfer in cases where tasks interfere. We test our algorithm against two state-of-the-art multi-task/transfer learning approaches and show comparable or superior performance while requiring fewer network parameters.

cs.LG↗

arXiv2019

Stabilization Time in Minority Processes

We analyze the stabilization time of minority processes in graphs. A minority process is a dynamically changing coloring, where each node repeatedly changes its color to the color which is least frequent in its neighborhood. First, we present a simple $Ω(n^2)$ stabilization time lower bound in the sequential adversarial model. Our main contribution is a graph construction which proves a $Ω(n^{2-ε})$ stabilization time lower bound for any $ε>0$. This lower bound holds even if the order of nodes is chosen benevolently, not only in the sequential model, but also in any reasonable concurrent model of the process.

cs.DM↗

arXiv2019

Stabilization Time in Weighted Minority Processes

A minority process in a weighted graph is a dynamically changing coloring. Each node repeatedly changes its color in order to minimize the sum of weighted conflicts with its neighbors. We study the number of steps until such a process stabilizes. Our main contribution is an exponential lower bound on stabilization time. We first present a construction showing this bound in the adversarial sequential model, and then we show how to extend the construction to establish the same bound in the benevolent sequential model, as well as in any reasonable concurrent model. Furthermore, we show that the stabilization time of our construction remains exponential even for very strict switching conditions, namely, if a node only changes color when almost all (i.e., any specific fraction) of its neighbors have the same color. Our lower bound works in a wide range of settings, both for node-weighted and edge-weighted graphs, or if we restrict minority processes to the class of sparse graphs.

cs.DM↗

arXiv2018

Towards Secure and Efficient Payment Channels

Micropayment channels are the most prominent solution to the limitation on transaction throughput in current blockchain systems. However, in practice channels are risky because participants have to be online constantly to avoid fraud, and inefficient because participants have to open multiple channels and lock funds in them. To address the security issue, we propose a novel mechanism that involves watchtowers incentivized to watch the channels and reveal a fraud. Our protocol does not require participants to be online constantly watching the blockchain. The protocol is secure, incentive compatible and lightweight in communication. Furthermore, we present an adaptation of our protocol implementable on the Lightning protocol. Towards efficiency, we examine specific topological structures in the blockchain transaction graph and generalize the construction of channels to enable topologies better suited to specific real-world needs. In these cases, our construction reduces the required amount of signatures for a transaction and the total amount of locked funds in the system.

cs.CR↗