arXiv subjects

Aditya Sinha

Publications and source records attributed to Aditya Sinha.

14 recordsLinked to original sources

arXiv2026

Mult-DPO: Multinomial Direct Preference Optimization for Recommender Systems

Direct preference optimization (DPO) is a simple and effective alignment strategy for large language models (LLMs) based on pairwise preferences. In recommender systems, however, user feedback is rarely pairwise. For a given context, e.g., a user, a session, or a conversation, we typically observe set-wise preferences with multiple positive items, where every positive item should outrank every unobserved or explicitly negative item, with no prescribed order among the positives or the negatives themselves. A natural generalization is to use the Plackett-Luce (PL) reward model, which extends the Bradley-Terry reward model underlying vanilla DPO from pairwise preferences to full rankings of candidates. However, we show that adapting the PL model to set-wise preferences requires marginalizing over all positive orderings, where the resulting expression is combinatorial in complexity. To address this fundamental challenge, we propose Mult-DPO, a novel DPO objective with a tractable multinomial surrogate likelihood over set-wise preference events for the user-preference alignment of LLM-based recommender systems. The multinomial construction is not itself a ranking distribution, but it is defined on the same reward-induced weight space and admits a closed-form DPO-style objective, enabling direct alignment of LLMs with multiple candidates through a classification-style objective. In addition, we prove that the multinomial DPO loss is a tractable upper bound on the marginalized PL DPO loss when optimizing against the set-wise preference data. We further characterize the tightness of this bound in terms of the relative total weight of positives versus negatives, which provides insights into tightening the bound with richer or harder negatives. Finally, we extend Mult-DPO to the alignment of LLMs with multiple preference levels. Code is available at https://github.com/yaochenzhu/Mult_DPO

cs.IR↗

arXiv2026

Training Deliberative Monitors for Black-Box Scheming Detection

As autonomous agents become more capable of performing real-world tasks, distinguishing scheming behavior from benign task pursuit may become a central AI control problem. Existing monitors often rely on chain-of-thought access or internal activations, or use prompted frontier models, all of which can be unavailable, unreliable or expensive in deployment. In this work, we study action-only deliberative monitors: smaller open-weight models trained to detect scheming and sabotage from agentic trajectories without accessing the monitored agent's reasoning or model internals. Our method, inspired by deliberative alignment, uses a scheming specification to elicit structured rationales from a frontier teacher, filters them with a separate judge, and distills the highest-quality rationales into open-weight monitors with supervised fine-tuning and reinforcement learning. We train on five datasets, and evaluate across six out-of-distribution agentic misalignment benchmarks. We show that applying our method to Qwen3.5-27B yields higher performance than all low-cost frontier models as prompted monitors (Gemini 3.1 Flash-Lite, GPT-5.4 Nano, and Claude Haiku 4.5) and than Gemini 2.5 Pro, while also achieving lower marginal inference cost (token-metered USD per 1,000 evaluations). Stronger prompted frontier monitors (Gemini 3.1 Pro, GPT-5.4, Claude Sonnet 4.6, and Claude Opus 4.6) achieve higher performance but at roughly $16$--$34\times$ higher marginal inference cost. Several of our trained monitors are positioned on the empirical cost--performance Pareto frontier among the monitors we evaluate, providing practical low-cost, low-FPR alternatives to prompted frontier models.

cs.CL↗

arXiv2026

Beyond Continuity: Challenges of Context Switching in Multi-Turn Dialogue with LLMs

Users interacting with Large Language Models (LLMs) in a multi-turn conversation routinely refine their requests or pivot to new topics. LLMs, however, often miss these topic shifts and carry over irrelevant context from previous turns, leading to inaccurate responses. In this paper, we stress-test the multi-turn understanding of LLMs and study the following two sub-tasks: (1) detecting whether the user pivots or refines in the current turn, and (2) shortlisting relevant context from previous turns. To this end, we construct synthetic benchmarks based on real-world datasets from varied domains, as to simulate context shifts of different levels of difficulty. We then evaluate the zero-shot performance of ten LLMs (open-weight, closed-source and reasoning), and demonstrate that only some reasoning and strongly instructed LLMs are accurate in detecting pivots; open-weight LLMs struggle with the task and frequently carry stale context even with explicit cues; and all models suffer from a position bias. Based on the results, we discuss key takeaways for improving long-term robustness in multi-turn capabilities for LLMs.

cs.CL↗

arXiv2025

Experiences Building Enterprise-Level Privacy-Preserving Federated Learning to Power AI for Science

Federated learning (FL) is a promising approach to enabling collaborative model training without centralized data sharing, a crucial requirement in scientific domains where data privacy, ownership, and compliance constraints are critical. However, building user-friendly enterprise-level FL frameworks that are both scalable and privacy-preserving remains challenging, especially when bridging the gap between local prototyping and distributed deployment across heterogeneous client computing infrastructures. In this paper, based on our experiences building the Advanced Privacy-Preserving Federated Learning (APPFL) framework, we present our vision for an enterprise-grade, privacy-preserving FL framework designed to scale seamlessly across computing environments. We identify several key capabilities that such a framework must provide: (1) Scalable local simulation and prototyping to accelerate experimentation and algorithm design; (2) seamless transition from simulation to deployment; (3) distributed deployment across diverse, real-world infrastructures, from personal devices to cloud clusters and HPC systems; (4) multi-level abstractions that balance ease of use and research flexibility; and (5) comprehensive privacy and security through techniques such as differential privacy, secure aggregation, robust authentication, and confidential computing. We further discuss architectural designs to realize these goals. This framework aims to bridge the gap between research prototypes and enterprise-scale deployment, enabling scalable, reliable, and privacy-preserving AI for science.

cs.DC↗

arXiv2025

CADRE: Customizable Assurance of Data Readiness in Privacy-Preserving Federated Learning

Privacy-Preserving Federated Learning (PPFL) is a decentralized machine learning approach where multiple clients train a model collaboratively. PPFL preserves the privacy and security of a client's data without exchanging it. However, ensuring that data at each client is of high quality and ready for federated learning (FL) is a challenge due to restricted data access. In this paper, we introduce CADRE (Customizable Assurance of Data Readiness) for federated learning (FL), a novel framework that allows users to define custom data readiness (DR) metrics, rules, and remedies tailored to specific FL tasks. CADRE generates comprehensive DR reports based on the user-defined metrics, rules, and remedies to ensure datasets are prepared for FL while preserving privacy. We demonstrate a practical application of CADRE by integrating it into an existing PPFL framework. We conducted experiments across six datasets and addressed seven different DR issues. The results illustrate the versatility and effectiveness of CADRE in ensuring DR across various dimensions, including data quality, privacy, and fairness. This approach enhances the performance and reliability of FL models as well as utilizes valuable resources.

cs.CR↗

arXiv2025

FedCostAware: Enabling Cost-Aware Federated Learning on the Cloud

Federated learning (FL) is a distributed machine learning (ML) approach that allows multiple clients to collaboratively train ML models without exchanging original training data, offering a solution that is particularly valuable in sensitive domains such as biomedicine. However, training robust FL models often requires substantial computing resources from participating clients, which may not be readily available at institutions such as hospitals. While cloud platforms offer on-demand access to such resources, their usage can incur significant costs, particularly in distributed training scenarios where poor coordination strategies can lead to substantial resource wastage. To address this, we introduce FedCostAware, a cost-aware scheduling algorithm designed to optimize synchronous FL on cloud spot instances. FedCostAware addresses the challenges of training on spot instances and different client budgets by employing intelligent management of the lifecycle of spot instances. This approach minimizes resource idle time and overall expenses. Comprehensive experiments across multiple datasets demonstrate that FedCostAware significantly reduces cloud computing costs compared to conventional spot and on-demand schemes, enhancing the accessibility and affordability of FL.

cs.DC↗

arXiv2024

Learning Structured Representations with Hyperbolic Embeddings

Most real-world datasets consist of a natural hierarchy between classes or an inherent label structure that is either already available or can be constructed cheaply. However, most existing representation learning methods ignore this hierarchy, treating labels as permutation invariant. Recent work [Zeng et al., 2022] proposes using this structured information explicitly, but the use of Euclidean distance may distort the underlying semantic context [Chen et al., 2013]. In this work, motivated by the advantage of hyperbolic spaces in modeling hierarchical relationships, we propose a novel approach HypStructure: a Hyperbolic Structured regularization approach to accurately embed the label hierarchy into the learned representations. HypStructure is a simple-yet-effective regularizer that consists of a hyperbolic tree-based representation loss along with a centering loss, and can be combined with any standard task loss to learn hierarchy-informed features. Extensive experiments on several large-scale vision benchmarks demonstrate the efficacy of HypStructure in reducing distortion and boosting generalization performance especially under low dimensional scenarios. For a better understanding of structured representation, we perform eigenvalue analysis that links the representation geometry to improved Out-of-Distribution (OOD) detection performance seen empirically. The code is available at \url{https://github.com/uiuctml/HypStructure}.

cs.LG↗

arXiv2024

Matryoshka Representation Learning

Learned representations are a central component in modern ML systems, serving a multitude of downstream tasks. When training such representations, it is often the case that computational and statistical constraints for each downstream task are unknown. In this context rigid, fixed capacity representations can be either over or under-accommodating to the task at hand. This leads us to ask: can we design a flexible representation that can adapt to multiple downstream tasks with varying computational resources? Our main contribution is Matryoshka Representation Learning (MRL) which encodes information at different granularities and allows a single embedding to adapt to the computational constraints of downstream tasks. MRL minimally modifies existing representation learning pipelines and imposes no additional cost during inference and deployment. MRL learns coarse-to-fine representations that are at least as accurate and rich as independently trained low-dimensional representations. The flexibility within the learned Matryoshka Representations offer: (a) up to 14x smaller embedding size for ImageNet-1K classification at the same level of accuracy; (b) up to 14x real-world speed-ups for large-scale retrieval on ImageNet-1K and 4K; and (c) up to 2% accuracy improvements for long-tail few-shot classification, all while being as robust as the original representations. Finally, we show that MRL extends seamlessly to web-scale datasets (ImageNet, JFT) across various modalities -- vision (ViT, ResNet), vision + language (ALIGN) and language (BERT). MRL code and pretrained models are open-sourced at https://github.com/RAIVNLab/MRL.

cs.LG↗

arXiv2023

Programmable Nanowrinkle-Induced Room-Temperature Exciton Localization in Monolayer WSe2

Localized states in two-dimensional (2D) transition metal dichalcogenides (TMDCs) have been the subject of intense study, driven by potential applications in quantum information science. Despite the rapidly growing knowledge surrounding these emitters, their microscopic nature is still not fully understood, limiting their production and application. Motivated by this challenge, and by recent theoretical and experimental evidence showing that nanowrinkles generate localized room-temperature emitters, we demonstrate a method to intentionally induce wrinkles with collections of stressors, showing that long-range wrinkle direction and position are controllable with patterned array design. Nano-photoluminescence (nano-PL) imaging combined with detailed strain modeling based on measured wrinkle topography establishes a correlation between wrinkle properties, particularly shear strain, and localized exciton emission. Beyond the array-induced super-wrinkles, nano-PL spatial maps further reveal that the strain environment around individual stressors is heterogeneous due to the presence of fine wrinkles that are less deterministic. Detailed nanoscale hyperspectral images uncover a wide range of low-energy emission peaks originating from these fine wrinkles, and show that the states can be tightly confined to regions < 10 nm, even in ambient conditions. These results establish a promising potential route towards realizing room temperature quantum emission in 2D TMDC systems.

cond-mat.mes-hall↗

arXiv2022

Node-Level Differentially Private Graph Neural Networks

Graph Neural Networks (GNNs) are a popular technique for modelling graph-structured data and computing node-level representations via aggregation of information from the neighborhood of each node. However, this aggregation implies an increased risk of revealing sensitive information, as a node can participate in the inference for multiple nodes. This implies that standard privacy-preserving machine learning techniques, such as differentially private stochastic gradient descent (DP-SGD) - which are designed for situations where each data point participates in the inference for one point only - either do not apply, or lead to inaccurate models. In this work, we formally define the problem of learning GNN parameters with node-level privacy, and provide an algorithmic solution with a strong differential privacy guarantee. We employ a careful sensitivity analysis and provide a non-trivial extension of the privacy-by-amplification technique to the GNN setting. An empirical evaluation on standard benchmark datasets demonstrates that our method is indeed able to learn accurate privacy-preserving GNNs which outperform both private and non-private methods that completely ignore graph information.

cs.LG↗

arXiv2022

IGLU: Efficient GCN Training via Lazy Updates

Training multi-layer Graph Convolution Networks (GCN) using standard SGD techniques scales poorly as each descent step ends up updating node embeddings for a large portion of the graph. Recent attempts to remedy this sub-sample the graph that reduces compute but introduce additional variance and may offer suboptimal performance. This paper develops the IGLU method that caches intermediate computations at various GCN layers thus enabling lazy updates that significantly reduce the compute cost of descent. IGLU introduces bounded bias into the gradients but nevertheless converges to a first-order saddle point under standard assumptions such as objective smoothness. Benchmark experiments show that IGLU offers up to 1.2% better accuracy despite requiring up to 88% less compute.

cs.LG↗

arXiv2021

Tensor spectrum of turbulence-sourced gravitational waves as a constraint on graviton mass

We consider a generic dispersive massive gravity theory and numerically study its resulting modified energy and strain spectra of tensor gravitational waves (GWs) sourced by (i) fully developed turbulence during the electroweak phase transition (EWPT) and (ii) forced hydromagnetic turbulence during the QCD phase transition (QCDPT). The GW spectra are then computed in both spatial and temporal Fourier domains. We find, from the spatial spectra, that the slope modifications are weakly dependent on the eddy size at QCDPT, and, from the temporal spectra, that the modifications are pronounced in the $1$--$10{\rm nHz}$ range -- the sensitivity range of the North American Nanohertz Observatory for Gravitational Waves (NANOGrav) -- for a graviton mass $m_{\rm g}$ in the range $2\times10^{-23}{\rm eV}\lesssim m_{\rm g}c^2\lesssim7\times10^{-22}{\rm eV}$.

gr-qc↗

arXiv2021

Non-Lorentzian Chaos and Cosmological Holography

We study chaos in non-Lorentzian field theories, specifically Galilean and Carrollian conformal field theories in two dimensions. In a large central charge limit, we find that the Lyapunov exponent saturates the bound on chaos, conjectured originally for relativistic field theories. We recover the same Lyapunov exponent holographically by a shock-wave calculation in three-dimensional flat space cosmologies, providing further evidence for flat space holography.

hep-th↗

arXiv2020

Rich-Item Recommendations for Rich-Users: Exploiting Dynamic and Static Side Information

In this paper, we study the problem of recommendation system where the users and items to be recommended are rich data structures with multiple entity types and with multiple sources of side-information in the form of graphs. We provide a general formulation for the problem that captures the complexities of modern real-world recommendations and generalizes many existing formulations. In our formulation, each user/document that requires a recommendation and each item or tag that is to be recommended, both are modeled by a set of static entities and a dynamic component. The relationships between entities are captured by several weighted bipartite graphs. To effectively exploit these complex interactions and learn the recommendation model, we propose MEDRES- a multiple graph-CNN based novel deep-learning architecture. MEDRES uses AL-GCN, a novel graph convolution network block, that harnesses strong representative features from the underlying graphs. Moreover, in order to capture highly heterogeneous engagement of different users with the system and constraints on the number of items to be recommended, we propose a novel ranking metric pAp@k along with a method to optimize the metric directly. We demonstrate effectiveness of our method on two benchmarks: a) citation data, b) Flickr data. In addition, we present two real-world case studies of our formulation and the MEDRES architecture. We show how our technique can be used to naturally model the message recommendation problem and the teams recommendation problem in the Microsoft Teams (MSTeams) product and demonstrate that it is 5-6% points more accurate than the production-grade models.

cs.LG↗