Rishi Sharma is a Computer Science PhD researcher at EPFL specializing in privacy-preserving decentralized learning systems. His research combines distributed systems, machine learning, and privacy to develop collaborative AI frameworks that are scalable, resource-efficient, and privacy-aware.https://rishisharma.netlify.app/https://rishisharma.netlify.app/publication/sharma-2026-cesar/https://rishisharma.netlify.app/publication/sharma-2026-cesar/Decentralized learning (DL) enables participants to collaboratively train models without a central server, yet it faces significant scalability challenges that demand sparsification to reduce the prohibitive communication costs of peer-to-peer exchange. While secure aggregation effectively mitigates privacy risks in standard settings, it has remained fundamentally incompatible with sparsification in decentralized networks due to the mismatch of indices across local updates, forcing a trade-off between communication efficiency and privacy. This paper introduces CESAR, a novel protocol that resolves this incompatibility by integrating secure aggregation and sparsification to provide provable defense against honest-but-curious and colluding adversaries. By coordinating masks over parameter intersections, CESAR supports node dropouts and robust privacy without central aggregation. Empirical evaluations on models up to 124 million parameters demonstrate that CESAR matches the accuracy of non-private baselines under equal sparsification while cutting total data exchange by 66.7% compared to a standard full-parameter decentralized protocol (D-PSGD). With TopK sparsification on IID data, CESAR even exceeds by 0.3% the accuracy achieved by D-PSGD with sparsification. Collectively, these results establish CESAR as the first decentralized protocol to achieve both privacy and communication efficiency through secure aggregation in DL.Tue, 14 Jul 2026 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2026-webofagents/https://rishisharma.netlify.app/publication/sharma-2026-webofagents/Collaborative agentic AI is projected to transform entire industries by enabling AI-powered agents to autonomously perceive, plan, and act within digital environments. Yet, current solutions in this field are all built in isolation, and we are rapidly heading toward a landscape of fragmented, incompatible ecosystems. In this position paper, we argue that interoperability, achieved by the adoption of minimal standards, is essential to ensure open, secure, web-scale, and widely-adopted agentic ecosystems. To this end, we devise a minimal architectural foundation for collaborative agentic AI, named Web of Agents, which is composed of four components: agent-to-agent messaging, interaction interoperability, state management, and agent discovery. Web of Agents adopts existing standards and reuses existing infrastructure where possible. With Web of Agents, we take a first but critical step toward interoperable agentic systems and offer a pragmatic path forward before ecosystem fragmentation becomes the norm.Thu, 30 Apr 2026 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2026-orbit/https://rishisharma.netlify.app/publication/sharma-2026-orbit/Large language models (LLMs) are increasingly deployed as autonomous agents that execute complex tasks through long sequences of reasoning steps and tool calls. Serving these agentic workloads at scale is a growing priority, yet existing LLM inference engines are mostly designed for single-pass, chat-style interactions. Agentic tasks, however, can fail, and under system congestion, policies such as first come first serve (FCFS) allocate GPU resources equally to all tasks regardless of their likelihood of success, letting failing tasks inflate queue waiting times for tasks that will complete correctly. We introduce Orbit, a trajectory-aware scheduler that addresses this by dynamically deprioritizing tasks predicted to fail, shifting resources toward likely-correct tasks. We present two scheduling algorithms. Orbit-step is a lightweight heuristic that assigns priority inversely proportional to the number of steps a task has taken, leveraging the empirical observation that longer-running tasks are more likely to fail. Orbit-judge uses an asynchronous LLM-as-a-judge that reads partial execution trajectories and predicts task correctness online, enabling finer-grained prioritization. We evaluate both variants on the GAIA benchmark using the Magentic-One benchmark and the GPT-OSS-120B model. Compared to vLLM's default FCFS policy, Orbit-step and Orbit-judge show 11.2% and 4.3% decrease in average end-to-end latency, respectively, for correct tasks. These results demonstrate that real-time trajectory observation is a practical and effective signal for resource allocation in agentic serving systems.Mon, 27 Apr 2026 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2026-sonar/https://rishisharma.netlify.app/publication/sharma-2026-sonar/Decentralized machine learning relies on peer-to-peer communication, yet the role of network topology in shaping learning dynamics remains poorly understood due to the lack of controlled, reproducible evaluation frameworks. We present SONAR, a modular framework for topology-aware decentralized learning that unifies communication, topology management, and fine-grained telemetry, enabling end-to-end measurement of performance, communication, robustness, and privacy under consistent conditions. Using SONAR, we show that topology is a first-class systems variable whose impact amplifies with scale and data heterogeneity. We find that sparse, structured topologies (e.g., rings and tori) can achieve comparable or superior accuracy to dense graphs at substantially lower communication cost under circumstances, revealing a clear efficiency frontier. We further identify and provide insights on collaborator collapse, a systematic failure mode in adaptive collaboration, where similarity-based neighbor selection reduces diversity and degrades generalization. By exposing topology as a controllable and measurable dimension, SONAR enables systematic, reproducible evaluation of decentralized learning and provides practical guidance for designing efficient and robust collaborative systems.Tue, 27 Jan 2026 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2026-agentplanning/https://rishisharma.netlify.app/publication/sharma-2026-agentplanning/Indirect prompt injection attacks threaten AI agents that execute consequential actions, motivating deterministic system-level defenses. Such defenses can provably block unsafe actions by enforcing confidentiality and integrity policies, but currently appear costly: they reduce task completion rates and increase token usage compared to probabilistic defenses. We argue that existing evaluations miss a key benefit of system-level defenses: reduced reliance on human oversight. We introduce autonomy metrics to quantify this benefit: the fraction of consequential actions an agent can execute without human-in-the-loop (HITL) approval while preserving security. To increase autonomy, we design a security-aware agent that (i) introduces richer HITL interactions, and (ii) explicitly plans for both task progress and policy compliance. We implement this agent design atop an existing information-flow control defense against prompt injection and evaluate it on the AgentDojo and WASP benchmarks. Experiments show that this approach yields higher autonomy without sacrificing utility (task completion).Mon, 26 Jan 2026 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-harmoeny/https://rishisharma.netlify.app/publication/sharma-2025-harmoeny/Mixture-of-Experts (MoE) models offer computational efficiency during inference by activating only a subset of specialized experts for a given input. This enables efficient model scaling on multi-GPU systems that use expert parallelism without compromising performance. However, load imbalance among experts and GPUs introduces waiting times, which can significantly increase inference latency. To address this challenge, we propose HarMoEny, a novel solution to address MoE load imbalance through two simple techniques: (i) dynamic token redistribution to underutilized GPUs and (ii) asynchronous prefetching of experts from the system to GPU memory. These techniques achieve a near-perfect load balance among experts and GPUs and mitigate delays caused by overloaded GPUs. We implement HarMoEny and compare its latency and throughput with four MoE baselines using real-world and synthetic datasets. Under heavy load imbalance, HarMoEny increases throughput by 37%-70% and reduces time-to-first-token by 34%-41%, compared to the next-best baseline. Moreover, our ablation study demonstrates that HarMoEny's scheduling policy reduces the GPU idling time by up to 84% compared to the baseline policies.Thu, 01 Jan 2026 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-sonarweb/https://rishisharma.netlify.app/publication/sharma-2025-sonarweb/Most federated learning (FL) frameworks assume reliable networks and homogeneous devices, limiting their applicability in mobile and edge environments where connectivity is intermittent and devices are highly heterogeneous. We introduce SONAR Web, an open-source framework for fully decentralized, cross-platform collaborative learning between browsers, servers, tablets, and smartphones. SONAR Web decouples the learning protocol from the underlying client platform through a platform-agnostic configuration interface—enabling Python, JavaScript, and mobile clients to seamlessly interoperate in real time. By combining peer-to-peer RTC protocols with communication-efficient techniques from FL, SONAR Web supports privacy-preserving training without centralized orchestration. We demonstrate SONAR Web's robustness through deployments on real-world devices and networks, showing resilience under heterogeneous network conditions and resource variability. SONAR Web provides a unified, language-agnostic interface for decentralized learning, enabling seamless collaboration across heterogeneous devices and runtimes—advancing scalable, inclusive, and real-time model training at the mobile and edge frontier.Tue, 04 Nov 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-boostinggel/https://rishisharma.netlify.app/publication/sharma-2025-boostinggel/Federated learning (FL) enables a set of client devices to collaboratively train a model without sharing raw data. This process, though, operates under the constrained computation and communication resources of edge devices. These constraints combined with systems heterogeneity force some participating clients to perform fewer local updates than expected by the server, thus slowing down convergence. Exhaustive tuning of hyperparameters in FL, furthermore, can be resource-intensive, without which the convergence is adversely affected. In this work, we propose GEL, the guess and learn algorithm. GEL enables constrained edge devices to perform additional learning through guessed updates on top of gradient-based steps. These guesses are gradientless, i.e., participating clients leverage them for free. Our generic guessing algorithm (i) can be flexibly combined with several state-of-the-art algorithms including FedProx + GeL, FedNova, FedYogi or ScaleFL; and (ii) achieves significantly improved performance when the learning rates are not best tuned. We conduct extensive experiments and show that GEL can boost empirical convergence by up to 40% in resource-constrained networks while relieving the need for exhaustive learning rate tuning.Tue, 10 Jun 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-noiseless/https://rishisharma.netlify.app/publication/sharma-2025-noiseless/Decentralized learning (DL) enables collaborative learning without a server and without training data leaving the users' devices. However, the models shared in DL can still be used to infer training data. Conventional privacy defenses such as differential privacy and secure aggregation fall short in effectively safeguarding user privacy in DL. We introduce Shatter, a novel DL approach in which nodes create virtual nodes (VNs) to disseminate chunks of their full model on their behalf. This enhances privacy by (i) preventing attackers from collecting full models from other nodes, and (ii) hiding the identity of the original node that produced a given model chunk. We theoretically prove the convergence of Shatter and provide a formal analysis demonstrating how Shatter reduces the efficacy of attacks compared to when exchanging full models between participating nodes. We evaluate the convergence and attack resilience of Shatter with existing DL algorithms, with heterogeneous datasets, and against three standard privacy attacks, including gradient inversion. Our evaluation shows that Shatter not only renders these privacy attacks infeasible when each node operates 16 VNs but also exhibits a positive impact on model convergence compared to standard DL. This enhanced privacy comes with a manageable increase in communication volume.Thu, 01 May 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-low/https://rishisharma.netlify.app/publication/sharma-2025-low/Decentralized learning (DL) is an emerging paradigm of collaborative machine learning that enables nodes in a network to train models collectively without sharing their raw data or relying on a central server. This paper introduces Zip-DL, a privacy-aware DL algorithm that leverages correlated noise to achieve robust privacy against local adversaries while ensuring efficient convergence at low communication costs. By progressively neutralizing the noise added during distributed averaging, Zip-DL combines strong privacy guarantees with high model accuracy. Its design requires only one communication round per gradient descent iteration, significantly reducing communication overhead compared to competitors. We establish theoretical bounds on both convergence speed and privacy guarantees. Moreover, extensive experiments demonstrating Zip-DL’s practical applicability make it outperform state-of-the-art methods in the accuracy vs. vulnerability trade-off. Specifically, Zip-DL (i) reduces membership-inference attack success rates by up to 35% compared to baseline DL, (ii) decreases attack efficacy by up to 13% compared to competitors offering similar utility, and (iii) achieves up to 59% higher accuracy to completely nullify a basic attack scenario, compared to a state-of-the-art privacy-preserving approach under the same threat model. These results position Zip-DL as a practical and efficient solution for privacy-preserving decentralized learning in real-world applications.Wed, 30 Apr 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-boosting/https://rishisharma.netlify.app/publication/sharma-2025-boosting/Decentralized learning (DL) is an emerging technique that allows nodes on the web to collaboratively train machine learning models without sharing raw data. Dealing with stragglers, i.e., nodes with slower compute or communication than others, is a key challenge in DL. We present DivShare, a novel asynchronous DL algorithm that achieves fast model convergence in the presence of communication stragglers. DivShare achieves this by having nodes fragment their models into parameter subsets and send, in parallel to computation, each subset to a random sample of other nodes instead of sequentially exchanging full models. The transfer of smaller fragments allows more efficient usage of the collective bandwidth and enables nodes with slow network links to quickly contribute with at least some of their model parameters. By theoretically proving the convergence of DivShare, we provide, to the best of our knowledge, the first formal proof of convergence for a DL algorithm that accounts for the effects of asynchronous communication with delays. We experimentally evaluate DivShare against two state-of-the-art DL baselines, AD-PSGD and Swift, and with two standard datasets, CIFAR-10 and MovieLens. We find that DivShare with communication stragglers lowers time-to-accuracy by up to 3.9x compared to AD-PSGD on the CIFAR-10 dataset. Compared to baselines, DivShare also achieves up to 19.4% better accuracy and 9.5% lower test loss on the CIFAR-10 and MovieLens datasets, respectively.Tue, 01 Apr 2025 00:00:00 GMThttps://rishisharma.netlify.app/post/event-2025-decentralizehack/https://rishisharma.netlify.app/post/event-2025-decentralizehack/Collaboration 🤝🏽 vs. ⚔️ competition!Mon, 24 Feb 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-fair/https://rishisharma.netlify.app/publication/sharma-2025-fair/Decentralized learning (DL) is an emerging approach that enables nodes to collaboratively train a machine learning model without sharing raw data. In many application domains, such as healthcare, this approach faces challenges due to the high level of heterogeneity in the training data's feature space. Such feature heterogeneity lowers model utility and negatively impacts fairness, particularly for nodes with under-represented training data. In this paper, we introduce Facade, a clustering-based DL algorithm specifically designed for fair model training when the training data exhibits several distinct features. The challenge of Facade is to assign nodes to clusters, one for each feature, based on the similarity in the features of their local data, without requiring individual nodes to know apriori which cluster they belong to. Facade (1) dynamically assigns nodes to their appropriate clusters over time, and (2) enables nodes to collaboratively train a specialized model for each cluster in a fully decentralized manner. We theoretically prove the convergence of Facade, implement our algorithm, and compare it against three state-of-the-art baselines. Our experimental results on three datasets demonstrate the superiority of our approach in terms of model accuracy and fairness compared to all three competitors. Compared to the best-performing baseline, Facade on the CIFAR-10 dataset also reduces communication costs by 32.3% to reach a target accuracy when cluster sizes are imbalanced.Sat, 15 Feb 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-practical/https://rishisharma.netlify.app/publication/sharma-2025-practical/Federated Learning (FL) enables end-user devices to collaboratively train ML models without sharing raw data, thereby preserving data privacy. In FL, a central parameter server coordinates the learning process by iteratively aggregating the trained models received from clients. Yet, deploying a central server is not always feasible due to hardware unavailability, infrastructure constraints, or operational costs. We present Plexus, a fully decentralized FL system for large networks that operates without the drawbacks originating from having a central server. Plexus distributes the responsibilities of model aggregation and sampling among participating nodes while avoiding network-wide coordination. We evaluate Plexus using realistic traces for compute speed, pairwise latency and network capacity. Our experiments on three common learning tasks and with up to 1000 nodes empirically show that Plexus reduces time-to-accuracy by 1.4-1.6×, communication volume by 15.8-292× and training resources needed for convergence by 30.5-77.9× compared to conventional decentralized learning algorithms.Sat, 15 Feb 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2025-pyramidal/https://rishisharma.netlify.app/publication/sharma-2025-pyramidal/Analyzing gigapixel images is recognized as computationally demanding. In this paper, we introduce PyramidAI, a technique for analyzing gigapixel images with reduced computational cost. The proposed approach adopts a gradual analysis of the image, beginning with lower resolutions and progressively concentrating on regions of interest for detailed examination at higher resolutions. We investigated two strategies for tuning the accuracy-computation performance trade-off when implementing the adaptive resolution selection, validated against the Camelyon 16 dataset of biomedical images. Our results demonstrate that PyramidAI substantially decreases the amount of processed data required for analysis by up to 2.65×, while preserving the accuracy in identifying relevant sections on a single computer. To ensure democratization of gigapixel image analysis, we evaluated the potential to use mainstream computers to perform the computation by exploiting the parallelism potential of the approach. Using a simulator, we estimated the best data distribution and load balancing algorithm according to the number of workers. The selected algorithms were implemented and highlighted the same conclusions in a real-world setting. Analysis time is reduced from more than an hour to a few minutes using 12 modest workers, offering a practical solution for efficient large-scale image analysis.Wed, 01 Jan 2025 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2024-revisiting/https://rishisharma.netlify.app/publication/sharma-2024-revisiting/Federated Learning (FL) is an appealing approach to training machine learning models without sharing raw data. However, standard FL algorithms are iterative and thus induce a significant communication cost. One-Shot FL (OFL) trades the iterative exchange of models between clients and the server with a single round of communication, thereby saving substantially on communication costs. Not surprisingly, OFL exhibits a performance gap in terms of accuracy with respect to FL, especially under high data heterogeneity. We introduce Fens, a novel federated ensembling scheme that approaches the accuracy of FL with the communication efficiency of OFL. Learning in Fens proceeds in two phases: first, clients train models locally and send them to the server, similar to OFL; second, clients collaboratively train a lightweight prediction aggregator model using FL. We showcase the effectiveness of Fens through exhaustive experiments spanning several datasets and heterogeneity levels. In the particular case of heterogeneously distributed CIFAR-10 dataset, Fens achieves up to a 26.9% higher accuracy over SOTA OFL, being only 3.1% lower than FL. At the same time, Fens incurs at most 4.3x more communication than OFL, whereas FL is at least 10.9x more communication-intensive than Fens.Tue, 10 Dec 2024 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2024-energy/https://rishisharma.netlify.app/publication/sharma-2024-energy/SkipTrain is a novel Decentralized Learning (DL) algorithm, which minimizes energy consumption in decentralized learning by strategically skipping some training rounds and substituting them with synchronization rounds. These training-silent periods, besides saving energy, also allow models to better mix and produce models with superior accuracy than typical DL algorithms. Our empirical evaluations with 256 nodes demonstrate that SkipTrain reduces energy consumption by 50% and increases model accuracy by up to 12% compared to D-PSGD, the conventional DL algorithm.Wed, 01 May 2024 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2023-epidemic/https://rishisharma.netlify.app/publication/sharma-2023-epidemic/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 1.6x quicker than baseline DL algorithms and attains 1.8% higher accuracy for the same communication volume.Fri, 01 Dec 2023 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2023-get/https://rishisharma.netlify.app/publication/sharma-2023-get/Decentralized learning (DL) systems have been gaining popularity because they avoid raw data sharing by communicating only model parameters, hence preserving data confidentiality. However, the large size of deep neural networks poses a significant challenge for decentralized training, since each node needs to exchange gigabytes of data, overloading the network. In this paper, we address this challenge with JWINS, a communication-efficient and fully decentralized learning system that shares only a subset of parameters through sparsification. JWINS uses wavelet transform to limit the information loss due to sparsification and a randomized communication cut-off that reduces communication usage without damaging the performance of trained models. We demonstrate empirically with 96 DL nodes on non-IID datasets that JWINS can achieve similar accuracies to full-sharing DL while sending up to 64% fewer bytes. Additionally, on low communication budgets, JWINS outperforms the state-of-the-art communication-efficient DL algorithm Choco-SGD by up to 4x in terms of network savings and time.Sat, 01 Jul 2023 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2023-decentralized/https://rishisharma.netlify.app/publication/sharma-2023-decentralized/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.Sat, 01 Apr 2023 00:00:00 GMThttps://rishisharma.netlify.app/publication/sharma-2022-zs3/https://rishisharma.netlify.app/publication/sharma-2022-zs3/With the advent of multi-core systems, GPUs and FPGAs, loop parallelization has become a promising way to speed-up program execution. Correspondingly, researchers have developed techniques to parallelize loops that do not carry dependences across iterations, and/or call pure functions. However, in languages such as Java with managed runtimes, it is practically infeasible to perform com­plex dependence analysis during JIT compilation. In this paper, we propose 2S3, a first of its kind loop parallelizer for Java programs that marks parallelizable loops for heterogeneous architectures using TornadoVM (a Graal-based VM that supports insertion of @Parallel constructs for loop parallelization).ZS3 statically performs dependence and purity analysis of Java programs in the Soot framework, to generate constraints under which a given loop can be parallelized. These constraints are fed to the 23 theorem prover (which we have integrated with Soot) to annotate parallelizable loops with the @Parallel construct. We have also added runtime support in TornadoVM to use static anal­ysis results for loop parallelization. Our evaluation over standard parallelization kernels shows that 2S3 correctly parallelizes 61.3% of manually parallelizable loops, with an efficient static analysis and a near-zero runtime overhead. 2S3 is not only the first tool that per­forms program-analysis based parallelization for a real-world JVM, but also the first to integrate 23 with Soot for loop parallelization.Tue, 01 Nov 2022 00:00:00 GMThttps://rishisharma.netlify.app/post/blog-2021-gradapp/https://rishisharma.netlify.app/post/blog-2021-gradapp/Some important tips and resources for Grad School Applications.Thu, 14 Oct 2021 00:00:00 GMThttps://rishisharma.netlify.app/post/life-2021-epfl/https://rishisharma.netlify.app/post/life-2021-epfl/The Journey from IIT Mandi to EPFL.Wed, 01 Sep 2021 00:00:00 GMThttps://rishisharma.netlify.app/post/blog-2020-differentiation/https://rishisharma.netlify.app/post/blog-2020-differentiation/Using DAGs to find the derivatives of complex functions.Sat, 09 May 2020 00:00:00 GMT