arXiv subjects

Sriram Vishwanath

Publications and source records attributed to Sriram Vishwanath.

At least 37 records · Page 2Linked to original sources

arXiv2026

Minimax Rates for Hyperbolic Hierarchical Learning

We prove an exponential separation in sample complexity between Euclidean and hyperbolic representations for learning on hierarchical data under standard Lipschitz regularization. For depth-$R$ hierarchies with branching factor $m$, we first establish a geometric obstruction for Euclidean space: any bounded-radius embedding forces volumetric collapse, mapping exponentially many tree-distant points to nearby locations. This necessitates Lipschitz constants scaling as $\exp(Ω(R))$ to realize even simple hierarchical targets, yielding exponential sample complexity under capacity control. We then show this obstruction vanishes in hyperbolic space: constant-distortion hyperbolic embeddings admit $O(1)$-Lipschitz realizability, enabling learning with $n = O(mR \log m)$ samples. A matching $Ω(mR \log m)$ lower bound via Fano's inequality establishes that hyperbolic representations achieve the information-theoretic optimum. We also show a geometry-independent bottleneck: any rank-$k$ prediction space captures only $O(k)$ canonical hierarchical contrasts.

stat.ML↗

arXiv2026

OPTIMUM-DERAM: Highly Consistent, Scalable, and Secure Multi-Object Memory using RLNC

This paper introduces OPTIMUM-DERAM, a highly consistent, scalable, secure, and decentralized shared memory solution. Traditional distributed shared memory implementations offer multi-object support by multi-threading a single object memory instance over the same set of data hosts. While theoretically sound, the amount of resources required made such solutions prohibitively expensive in practical systems. OPTIMUM-DERAM proposes a decentralized, reconfigurable, atomic read/write shared memory (DeRAM) that: (i) achieves improved performance and storage scalability by leveraging Random Linear Network Codes (RLNC); (ii) scales in the number of supported atomic objects by introducing a new object placement and discovery approach based on a consistent hashing ring; (iii) scales in the number of participants by allowing dynamic joins and departures leveraging a blockchain oracle to serve as a registry service; and (iv) is secure against malicious behavior by tolerating Byzantine failures. Experimental results over a globally distributed set of nodes, help us realize the performance and scalability gains of OPTIMUM-DERAM over previous distributed shared memory solutions (i.e., the ABD algorithm [3])

cs.DC↗

arXiv2026

Mean Field Analysis of Blockchain Systems

We present a novel framework for analyzing blockchain consensus mechanisms by modeling blockchain growth as a Partially Observable Stochastic Game (POSG) which we reduce to a set of Partially Observable Markov Decision Processes (POMDPs) through the use of the mean field approximation. This approach formalizes the decision-making process of miners in Proof-of-Work (PoW) systems and enables a principled examination of block selection strategies as well as steady state analysis of the induced Markov chain. By leveraging a mean field game formulation, we efficiently characterize the information asymmetries that arise in asynchronous blockchain networks. Our first main result is an exact characterization of the tradeoff between network delay and PoW efficiency--the fraction of blocks which end up in the longest chain. We demonstrate that the tradeoff observed in our model at steady state aligns closely with theoretical findings, validating our use of the mean field approximation. Our second main result is a rigorous equilibrium analysis of the Longest Chain Rule (LCR). We show that the LCR is a mean field equilibrium and that it is uniquely optimal in maximizing PoW efficiency under certain mild assumptions. This result provides the first formal justification for continued use of the LCR in decentralized consensus protocols, offering both theoretical validation and practical insights. Beyond these core results, our framework supports flexible experimentation with alternative block selection strategies, system dynamics, and reward structures. It offers a systematic and scalable substitute for expensive test-net deployments or ad hoc analysis. While our primary focus is on Nakamoto-style blockchains, the model is general enough to accommodate other architectures through modifications to the underlying MDP.

cs.GT↗

arXiv2025

WavShape: Information-Theoretic Speech Representation Learning for Fair and Privacy-Aware Audio Processing

Speech embeddings often retain sensitive attributes such as speaker identity, accent, or demographic information, posing risks in biased model training and privacy leakage. We propose WavShape, an information-theoretic speech representation learning framework that optimizes embeddings for fairness and privacy while preserving task-relevant information. We leverage mutual information (MI) estimation using the Donsker-Varadhan formulation to guide an MI-based encoder that systematically filters sensitive attributes while maintaining speech content essential for downstream tasks. Experimental results on three known datasets show that WavShape reduces MI between embeddings and sensitive attributes by up to 81% while retaining 97% of task-relevant information. By integrating information theory with self-supervised speech models, this work advances the development of fair, privacy-aware, and resource-efficient speech systems.

cs.SD↗

arXiv2025

PhonemeFake: Redefining Deepfake Realism with Language-Driven Segmental Manipulation and Adaptive Bilevel Detection

Deepfake (DF) attacks pose a growing threat as generative models become increasingly advanced. However, our study reveals that existing DF datasets fail to deceive human perception, unlike real DF attacks that influence public discourse. It highlights the need for more realistic DF attack vectors. We introduce PhonemeFake (PF), a DF attack that manipulates critical speech segments using language reasoning, significantly reducing human perception by up to 42% and benchmark accuracies by up to 94%. We release an easy-to-use PF dataset on HuggingFace and open-source bilevel DF segment detection model that adaptively prioritizes compute on manipulated regions. Our extensive experiments across three known DF datasets reveal that our detection model reduces EER by 91% while achieving up to 90% speed-up, with minimal compute overhead and precise localization beyond existing models as a scalable solution.

cs.CV↗

arXiv2025

Learnings from Scaling Visual Tokenizers for Reconstruction and Generation

Visual tokenization via auto-encoding empowers state-of-the-art image and video generative models by compressing pixels into a latent space. Although scaling Transformer-based generators has been central to recent advances, the tokenizer component itself is rarely scaled, leaving open questions about how auto-encoder design choices influence both its objective of reconstruction and downstream generative performance. Our work aims to conduct an exploration of scaling in auto-encoders to fill in this blank. To facilitate this exploration, we replace the typical convolutional backbone with an enhanced Vision Transformer architecture for Tokenization (ViTok). We train ViTok on large-scale image and video datasets far exceeding ImageNet-1K, removing data constraints on tokenizer scaling. We first study how scaling the auto-encoder bottleneck affects both reconstruction and generation -- and find that while it is highly correlated with reconstruction, its relationship with generation is more complex. We next explored the effect of separately scaling the auto-encoders' encoder and decoder on reconstruction and generation performance. Crucially, we find that scaling the encoder yields minimal gains for either reconstruction or generation, while scaling the decoder boosts reconstruction but the benefits for generation are mixed. Building on our exploration, we design ViTok as a lightweight auto-encoder that achieves competitive performance with state-of-the-art auto-encoders on ImageNet-1K and COCO reconstruction tasks (256p and 512p) while outperforming existing auto-encoders on 16-frame 128p video reconstruction for UCF-101, all with 2-5x fewer FLOPs. When integrated with Diffusion Transformers, ViTok demonstrates competitive performance on image generation for ImageNet-1K and sets new state-of-the-art benchmarks for class-conditional video generation on UCF-101.

cs.CV↗

arXiv2024

Improving DeFi Mechanisms with Dynamic Games and Optimal Control: A Case Study in Stablecoins

Stablecoins are a class of cryptocurrencies which aim at providing consistency and predictability, typically by pegging the token's value to that of a real world asset. Designing resilient decentralized stablecoins is a challenge, and prominent stablecoins today either (i) give up on decentralization, or (ii) rely on user-owned cryptocurrencies as collateral, exposing the token to exogenous price fluctuations. In this latter category, it is increasingly common to employ algorithmic mechanisms to automate risk management, helping maintain the peg. One example of this is Reflexer's RAI, which adapts its system-internal exchange rate (redemption price) to secondary market conditions according to a proportional control law. In this paper, we take this idea of active management a step further, and introduce a new kind of control scheme based on a Stackelberg game model between the token protocol and its users. By doing so, we show that (i) we can mitigate adverse depeg events that inevitably arise in a fixed-redemption scheme such as MakerDao's DAI and (ii) generally outperform a simpler, adaptive-redemption scheme such as RAI in the task of targeting a desired market price. We demonstrate these results through extensive simulations over a range of market conditions.

cs.GT↗

arXiv2024

Enhancing K-user Interference Alignment for Discrete Constellations via Learning

In this paper, we consider a K-user interference channel where interference among the users is neither too strong nor too weak, a scenario that is relatively underexplored in the literature. We propose a novel deep learning-based approach to design the encoder and decoder functions that aim to maximize the sumrate of the interference channel for discrete constellations. We first consider the MaxSINR algorithm, a state-of-the-art linear scheme for Gaussian inputs, as the baseline and then propose a modified version of the algorithm for discrete inputs. We then propose a neural network-based approach that learns a constellation mapping with the objective of maximizing the sumrate. We provide numerical results to show that the constellations learned by the neural network-based approach provide enhanced alignments, not just in beamforming directions but also in terms of the effective constellation at the receiver, thereby leading to improved sum-rate performance.

eess.SP↗

arXiv2024

Unified Auto-Encoding with Masked Diffusion

At the core of both successful generative and self-supervised representation learning models there is a reconstruction objective that incorporates some form of image corruption. Diffusion models implement this approach through a scheduled Gaussian corruption process, while masked auto-encoder models do so by masking patches of the image. Despite their different approaches, the underlying similarity in their methodologies suggests a promising avenue for an auto-encoder capable of both de-noising tasks. We propose a unified self-supervised objective, dubbed Unified Masked Diffusion (UMD), that combines patch-based and noise-based corruption techniques within a single auto-encoding framework. Specifically, UMD modifies the diffusion transformer (DiT) training process by introducing an additional noise-free, high masking representation step in the diffusion noising schedule, and utilizes a mixed masked and noised image for subsequent timesteps. By integrating features useful for diffusion modeling and for predicting masked patch tokens, UMD achieves strong performance in downstream generative and representation learning tasks, including linear probing and class-conditional generation. This is achieved without the need for heavy data augmentations, multiple views, or additional encoders. Furthermore, UMD improves over the computational efficiency of prior diffusion based methods in total training time. We release our code at https://github.com/philippe-eecs/small-vision.

cs.CV↗

arXiv2024

TexShape: Information Theoretic Sentence Embedding for Language Models

With the exponential growth in data volume and the emergence of data-intensive applications, particularly in the field of machine learning, concerns related to resource utilization, privacy, and fairness have become paramount. This paper focuses on the textual domain of data and addresses challenges regarding encoding sentences to their optimized representations through the lens of information-theory. In particular, we use empirical estimates of mutual information, using the Donsker-Varadhan definition of Kullback-Leibler divergence. Our approach leverages this estimation to train an information-theoretic sentence embedding, called TexShape, for (task-based) data compression or for filtering out sensitive information, enhancing privacy and fairness. In this study, we employ a benchmark language model for initial text representation, complemented by neural networks for information-theoretic compression and mutual information estimations. Our experiments demonstrate significant advancements in preserving maximal targeted information and minimal sensitive information over adverse compression ratios, in terms of predictive accuracy of downstream models that are trained using the compressed data.

cs.CL↗

arXiv2023

Majority is Not Required: A Rational Analysis of the Private Double-Spend Attack from a Sub-Majority Adversary

We study the incentives behind double-spend attacks on Nakamoto-style Proof-of-Work cryptocurrencies. In these systems, miners are allowed to choose which transactions to reference with their block, and a common strategy for selecting transactions is to simply choose those with the highest fees. This can be problematic if these transactions originate from an adversary with substantial (but less than 50\%) computational power, as high-value transactions can present an incentive for a rational adversary to attempt a double-spend attack if they expect to profit. The most common mechanism for deterring double-spend attacks is for the recipients of large transactions to wait for additional block confirmations (i.e., to increase the attack cost). We argue that this defense mechanism is not satisfactory, as the security of the system is contingent on the actions of its users. Instead, we propose that defending against double-spend attacks should be the responsibility of the miners; specifically, miners should limit the amount of transaction value they include in a block (i.e., reduce the attack reward). To this end, we model cryptocurrency mining as a mean-field game in which we augment the standard mining reward function to simulate the presence of a rational, double-spending adversary. We design and implement an algorithm which characterizes the behavior of miners at equilibrium, and we show that miners who use the adversary-aware reward function accumulate more wealth than those who do not. We show that the optimal strategy for honest miners is to limit the amount of value transferred by each block such that the adversary's expected profit is 0. Additionally, we examine Bitcoin's resilience to double-spend attacks. Assuming a 6 block confirmation time, we find that an attacker with at least 25% of the network mining power can expect to profit from a double-spend attack.

cs.GT↗

arXiv2023

Investigating Human-Identifiable Features Hidden in Adversarial Perturbations

Neural networks perform exceedingly well across various machine learning tasks but are not immune to adversarial perturbations. This vulnerability has implications for real-world applications. While much research has been conducted, the underlying reasons why neural networks fall prey to adversarial attacks are not yet fully understood. Central to our study, which explores up to five attack algorithms across three datasets, is the identification of human-identifiable features in adversarial perturbations. Additionally, we uncover two distinct effects manifesting within human-identifiable features. Specifically, the masking effect is prominent in untargeted attacks, while the generation effect is more common in targeted attacks. Using pixel-level annotations, we extract such features and demonstrate their ability to compromise target models. In addition, our findings indicate a notable extent of similarity in perturbations across different attack algorithms when averaged over multiple models. This work also provides insights into phenomena associated with adversarial perturbations, such as transferability and model interpretability. Our study contributes to a deeper understanding of the underlying mechanisms behind adversarial attacks and offers insights for the development of more resilient defense strategies for neural networks.

cs.LG↗

arXiv2023

iDML: Incentivized Decentralized Machine Learning

With the rising emergence of decentralized and opportunistic approaches to machine learning, end devices are increasingly tasked with training deep learning models on-devices using crowd-sourced data that they collect themselves. These approaches are desirable from a resource consumption perspective and also from a privacy preservation perspective. When the devices benefit directly from the trained models, the incentives are implicit - contributing devices' resources are incentivized by the availability of the higher-accuracy model that results from collaboration. However, explicit incentive mechanisms must be provided when end-user devices are asked to contribute their resources (e.g., computation, communication, and data) to a task performed primarily for the benefit of others, e.g., training a model for a task that a neighbor device needs but the device owner is uninterested in. In this project, we propose a novel blockchain-based incentive mechanism for completely decentralized and opportunistic learning architectures. We leverage a smart contract not only for providing explicit incentives to end devices to participate in decentralized learning but also to create a fully decentralized mechanism to inspect and reflect on the behavior of the learning architecture.

cs.LG↗

arXiv2023

Spatial and Statistical Modeling of Multi-Panel Millimeter Wave Self-Interference

Characterizing self-interference is essential to the design and evaluation of in-band full-duplex communication systems. Until now, little has been understood about this coupling in full-duplex systems operating at millimeter wave (mmWave) frequencies, and it has been shown that the highly-idealized models proposed for such do not align with practice. This work presents the first spatial and statistical model of mmWave self-interference backed by measurements, enabling engineers to draw realizations that exhibit the large-scale and small-scale spatial characteristics observed in our nearly 6.5 million measurements taken at 28 GHz. Core to our model is its use of system and model parameters having real-world meaning, which facilitates its extension to systems beyond our own phased array platform through proper parameterization. We demonstrate this by collecting nearly 13 million additional measurements to show that our model can generalize to two other system configurations. We assess our model by comparing it against actual measurements to confirm its ability to align spatially and in distribution with real-world self-interference. In addition, using both measurements and our model of self-interference, we evaluate an existing beamforming-based full-duplex mmWave solution to illustrate that our model can be reliably used to design new solutions and validate the performance improvements they may offer.

eess.SP↗

arXiv2022

Scalable Multi-Chain Coordination via the Hierarchical Longest Chain Rule

This paper introduces BlockReduce, a Proof-of-Work (PoW) based blockchain system which achieves high transaction throughput through a hierarchy of merged mined blockchains, each operating in parallel on a partition the overall application state. Most notably, the full PoW available within the network is applied to all blockchains in BlockReduce, and cross-blockchain state transitions are enabled seamlessly within the core protocol. This paper shows that, given a hierarchy of blockchains and its associated security model, the protocol scales superlinearly in transaction throughput with the number of blockchains operated by the protocol.

cs.DC↗

arXiv2022

A Control Theoretic Approach to Infrastructure-Centric Blockchain Tokenomics

There are a multitude of Blockchain-based physical infrastructure systems, operating on a crypto-currency enabled token economy, where infrastructure suppliers are rewarded with tokens for enabling, validating, managing and/or securing the system. However, today's token economies are largely designed without infrastructure systems in mind, and often operate with a fixed token supply (e.g., Bitcoin). This paper argues that token economies for infrastructure networks should be structured differently - they should continually incentivize new suppliers to join the network to provide services and support to the ecosystem. As such, the associated token rewards should gracefully scale with the size of the decentralized system, but should be carefully balanced with consumer demand to manage inflation and be designed to ultimately reach an equilibrium. To achieve such an equilibrium, the decentralized token economy should be adaptable and controllable so that it maximizes the total utility of all users, such as achieving stable (overall non-inflationary) token economies. Our main contribution is to model infrastructure token economies as dynamical systems - the circulating token supply, price, and consumer demand change as a function of the payment to nodes and costs to consumers for infrastructure services. Crucially, this dynamical systems view enables us to leverage tools from mathematical control theory to optimize the overall decentralized network's performance. Moreover, our model extends easily to a Stackelberg game between the controller and the nodes, which we use for robust, strategic pricing. In short, we develop predictive, optimization-based controllers that outperform traditional algorithmic stablecoin heuristics by up to $2.4 \times$ in simulations based on real demand data from existing decentralized wireless networks.

cs.DC↗

arXiv2022

Network Design for Social Welfare

In this paper, we consider the problem of network design on network games. We study the conditions on the adjacency matrix of the underlying network to design a game such that the Nash equilibrium coincides with the social optimum. We provide the examples for linear quadratic games that satisfy this condition. Furthermore, we identify conditions on properties of adjacency matrix that provide a unique solution using variational inequality formulation, and verify the robustness and continuity of the social cost under perturbations of the network. Finally we comment on individual rationality and extension of our results to large random networked games.

cs.GT↗

arXiv2022

STEER: Beam Selection for Full-Duplex Millimeter Wave Communication Systems

Modern millimeter wave (mmWave) communication systems rely on beam alignment to deliver sufficient beamforming gain to close the link between devices. We present a novel beam selection methodology for multi-panel, full-duplex mmWave systems, which we call STEER, that delivers high beamforming gain while significantly reducing the full-duplex self-interference coupled between the transmit and receive beams. STEER does not necessitate changes to conventional beam alignment methodologies nor additional over-the-air feedback, making it compatible with existing cellular standards. Instead, STEER uses conventional beam alignment to identify the general directions beams should be steered, and then it makes use of a minimal number of self-interference measurements to jointly select transmit and receive beams that deliver high gain in these directions while coupling low self-interference. We implement STEER on an industry-grade 28 GHz phased array platform and use further simulation to show that full-duplex operation with beams selected by STEER can notably outperform both half-duplex and full-duplex operation with beams chosen via conventional beam selection. For instance, STEER can reliably reduce self-interference by more than 20 dB and improve SINR by more than 10 dB, compared to conventional beam selection. Our experimental results highlight that beam alignment can be used not only to deliver high beamforming gain in full-duplex mmWave systems but also to mitigate self-interference to levels near or below the noise floor, rendering additional self-interference cancellation unnecessary with STEER.

eess.SP↗