arXiv subjects

Naman Agarwal

Publications and source records attributed to Naman Agarwal.

At least 37 records · Page 2Linked to original sources

arXiv2024

Adaptive Gradient Methods at the Edge of Stability

Very little is known about the training dynamics of adaptive gradient methods like Adam in deep learning. In this paper, we shed light on the behavior of these algorithms in the full-batch and sufficiently large batch settings. Specifically, we empirically demonstrate that during full-batch training, the maximum eigenvalue of the preconditioned Hessian typically equilibrates at a certain numerical value -- the stability threshold of a gradient descent algorithm. For Adam with step size $η$ and $β_1 = 0.9$, this stability threshold is $38/η$. Similar effects occur during minibatch training, especially as the batch size grows. Yet, even though adaptive methods train at the ``Adaptive Edge of Stability'' (AEoS), their behavior in this regime differs in a significant way from that of non-adaptive methods at the EoS. Whereas non-adaptive algorithms at the EoS are blocked from entering high-curvature regions of the loss landscape, adaptive gradient methods at the AEoS can keep advancing into high-curvature regions, while adapting the preconditioner to compensate. Our findings can serve as a foundation for the community's future understanding of adaptive gradient methods in deep learning.

cs.LG↗

arXiv2024

Towards Quantifying the Preconditioning Effect of Adam

There is a notable dearth of results characterizing the preconditioning effect of Adam and showing how it may alleviate the curse of ill-conditioning -- an issue plaguing gradient descent (GD). In this work, we perform a detailed analysis of Adam's preconditioning effect for quadratic functions and quantify to what extent Adam can mitigate the dependence on the condition number of the Hessian. Our key finding is that Adam can suffer less from the condition number but at the expense of suffering a dimension-dependent quantity. Specifically, for a $d$-dimensional quadratic with a diagonal Hessian having condition number $κ$, we show that the effective condition number-like quantity controlling the iteration complexity of Adam without momentum is $\mathcal{O}(\min(d, κ))$. For a diagonally dominant Hessian, we obtain a bound of $\mathcal{O}(\min(d \sqrt{d κ}, κ))$ for the corresponding quantity. Thus, when $d < \mathcal{O}(κ^p)$ where $p = 1$ for a diagonal Hessian and $p = 1/3$ for a diagonally dominant Hessian, Adam can outperform GD (which has an $\mathcal{O}(κ)$ dependence). On the negative side, our results suggest that Adam can be worse than GD for a sufficiently non-diagonal Hessian even if $d \ll \mathcal{O}(κ^{1/3})$; we corroborate this with empirical evidence. Finally, we extend our analysis to functions satisfying per-coordinate Lipschitz smoothness and a modified version of the Polyak-Łojasiewicz condition.

cs.LG↗

arXiv2023

Improved Differentially Private and Lazy Online Convex Optimization

We study the task of $(ε, δ)$-differentially private online convex optimization (OCO). In the online setting, the release of each distinct decision or iterate carries with it the potential for privacy loss. This problem has a long history of research starting with Jain et al. [2012] and the best known results for the regime of ε not being very small are presented in Agarwal et al. [2023]. In this paper we improve upon the results of Agarwal et al. [2023] in terms of the dimension factors as well as removing the requirement of smoothness. Our results are now the best known rates for DP-OCO in this regime. Our algorithms builds upon the work of [Asi et al., 2023] which introduced the idea of explicitly limiting the number of switches via rejection sampling. The main innovation in our algorithm is the use of sampling from a strongly log-concave density which allows us to trade-off the dimension factors better leading to improved results.

cs.CR↗

arXiv2023

HAVE-Net: Hallucinated Audio-Visual Embeddings for Few-Shot Classification with Unimodal Cues

Recognition of remote sensing (RS) or aerial images is currently of great interest, and advancements in deep learning algorithms added flavor to it in recent years. Occlusion, intra-class variance, lighting, etc., might arise while training neural networks using unimodal RS visual input. Even though joint training of audio-visual modalities improves classification performance in a low-data regime, it has yet to be thoroughly investigated in the RS domain. Here, we aim to solve a novel problem where both the audio and visual modalities are present during the meta-training of a few-shot learning (FSL) classifier; however, one of the modalities might be missing during the meta-testing stage. This problem formulation is pertinent in the RS domain, given the difficulties in data acquisition or sensor malfunctioning. To mitigate, we propose a novel few-shot generative framework, Hallucinated Audio-Visual Embeddings-Network (HAVE-Net), to meta-train cross-modal features from limited unimodal data. Precisely, these hallucinated features are meta-learned from base classes and used for few-shot classification on novel classes during the inference phase. The experimental results on the benchmark ADVANCE and AudioSetZSL datasets show that our hallucinated modality augmentation strategy for few-shot classification outperforms the classifier performance trained with the real multimodal information at least by 0.8-2%.

cs.CV↗

arXiv2023

Multi-User Reinforcement Learning with Low Rank Rewards

In this work, we consider the problem of collaborative multi-user reinforcement learning. In this setting there are multiple users with the same state-action space and transition probabilities but with different rewards. Under the assumption that the reward matrix of the $N$ users has a low-rank structure -- a standard and practically successful assumption in the offline collaborative filtering setting -- the question is can we design algorithms with significantly lower sample complexity compared to the ones that learn the MDP individually for each user. Our main contribution is an algorithm which explores rewards collaboratively with $N$ user-specific MDPs and can learn rewards efficiently in two key settings: tabular MDPs and linear MDPs. When $N$ is large and the rank is constant, the sample complexity per MDP depends logarithmically over the size of the state-space, which represents an exponential reduction (in the state-space size) when compared to the standard ``non-collaborative'' algorithms.

cs.LG↗

arXiv2023

Transient non-collinear magnetic state for all-optical magnetization switching

Resonant absorption of a photon by bound electrons in a solid can promote an electron to another orbital state or transfer it to a neighboring atomic site. Such a transition in a magnetically ordered material could affect the magnetic order. While this process is an obvious road map for optical control of magnetization, experimental demonstration of such a process remains challenging. Exciting a significant fraction of magnetic ions requires a very intense incoming light beam, as orbital resonances are often weak compared to above-band-gap excitations. In the latter case, a sizeable reduction of the magnetization occurs as the absorbed energy increases the spin temperature, masking the non-thermal optical effects. Here, using ultrafast x-ray spectroscopy, we were able to resolve changes in the magnetization state induced by resonant absorption of infrared photons in Co-doped yttrium iron garnet, with negligible thermal effects. We found that the optical excitation of the Co ions affects the two distinct magnetic Fe sublattices differently, resulting in a transient non-collinear magnetic state. The present results indicate that the all-optical magnetization switching most likely occurs due to the creation of a transient, non-collinear magnetic state followed by coherent spin rotations of the Fe moments.

cond-mat.mtrl-sci↗

arXiv2023

Variance-Reduced Conservative Policy Iteration

We study the sample complexity of reducing reinforcement learning to a sequence of empirical risk minimization problems over the policy space. Such reductions-based algorithms exhibit local convergence in the function space, as opposed to the parameter space for policy gradient algorithms, and thus are unaffected by the possibly non-linear or discontinuous parameterization of the policy class. We propose a variance-reduced variant of Conservative Policy Iteration that improves the sample complexity of producing a $\varepsilon$-functional local optimum from $O(\varepsilon^{-4})$ to $O(\varepsilon^{-3})$. Under state-coverage and policy-completeness assumptions, the algorithm enjoys $\varepsilon$-global optimality after sampling $O(\varepsilon^{-2})$ times, improving upon the previously established $O(\varepsilon^{-3})$ sample requirement.

cs.LG↗

arXiv2023

Ultrafast X-ray imaging of the light-induced phase transition in VO2

Using light to control transient phases in quantum materials is an emerging route to engineer new properties and functionality, with both thermal and non-thermal phases observed out of equilibrium. Transient phases are expected to be heterogeneous, either through photo-generated domain growth or by generating topological defects, and this impacts the dynamics of the system. However, this nanoscale heterogeneity has not been directly observed. Here we use time- and spectrally resolved coherent X-ray imaging to track the prototypical light induced insulator-to-metal phase transition in vanadium dioxide on the nanoscale with femtosecond time resolution. We show that the early-time dynamics are independent of the initial spatial heterogeneity and observe a 200 fs switch to the metallic phase. A heterogeneous response emerges only after hundreds of picoseconds. Through spectroscopic imaging, we reveal that the transient metallic phase is a highly orthorhombically strained rutile metallic phase, an interpretation that is in contrast to those based on spatially averaged probes. Our results demonstrate the critical importance of spatially and spectrally resolved measurements for understanding and interpreting the transient phases of quantum materials.

cond-mat.str-el↗

arXiv2023

Photon shot-noise limited transient absorption soft X-ray spectroscopy at the European XFEL

Femtosecond transient soft X-ray Absorption Spectroscopy (XAS) is a very promising technique that can be employed at X-ray Free Electron Lasers (FELs) to investigate out-of-equilibrium dynamics for material and energy research. Here we present a dedicated setup for soft X-rays available at the Spectroscopy & Coherent Scattering (SCS) instrument at the European X-ray Free Electron Laser (EuXFEL). It consists of a beam-splitting off-axis zone plate (BOZ) used in transmission to create three copies of the incoming beam, which are used to measure the transmitted intensity through the excited and unexcited sample, as well as to monitor the incoming intensity. Since these three intensity signals are detected shot-by-shot and simultaneously, this setup allows normalized shot-by-shot analysis of the transmission. For photon detection, the DSSC imaging detector, which is capable of recording up to 800 images at 4.5 MHz frame rate during the FEL burst, is employed and allows approaching the photon shot-noise limit. We review the setup and its capabilities, as well as the online and offline analysis tools provided to users.

physics.ins-det↗

arXiv2022

Best of Both Worlds in Online Control: Competitive Ratio and Policy Regret

We consider the fundamental problem of online control of a linear dynamical system from two different viewpoints: regret minimization and competitive analysis. We prove that the optimal competitive policy is well-approximated by a convex parameterized policy class, known as a disturbance-action control (DAC) policies. Using this structural result, we show that several recently proposed online control algorithms achieve the best of both worlds: sublinear regret vs. the best DAC policy selected in hindsight, and optimal competitive ratio, up to an additive correction which grows sublinearly in the time horizon. We further conclude that sublinear regret vs. the optimal competitive policy is attainable when the linear dynamical system is unknown, and even when a stabilizing controller for the dynamics is not available a priori.

cs.LG↗

arXiv2022

The interplay of local electron correlations and ultrafast spin dynamics in fcc Ni

The complex electronic structure of metallic ferromagnets is determined by a balance between exchange interaction, electron hopping leading to band formation, and local Coulomb repulsion. The interplay between the respective terms of the Hamiltonian is of fundamental interest, since it produces most, if not all, of the exotic phenomena observed in the solid state. By combining high energy and temporal resolution in femtosecond time-resolved X-ray absorption spectroscopy with ab initio time-dependent density functional theory we analyze the electronic structure in fcc Ni on the time scale of these interactions in a pump-probe experiment. We distinguish transient broadening and energy shifts in the absorption spectra, which we demonstrate to be caused by electron repopulation and correlation-induced modifications of the electronic structure, respectively. Importantly, the theoretical description of this experimental result hence requires to take the local Coulomb interaction into account, revealing a temporal interplay between band formation, exchange interaction, and Coulomb repulsion.

cond-mat.mtrl-sci↗

arXiv2022

Pushing the Efficiency-Regret Pareto Frontier for Online Learning of Portfolios and Quantum States

We revisit the classical online portfolio selection problem. It is widely assumed that a trade-off between computational complexity and regret is unavoidable, with Cover's Universal Portfolios algorithm, SOFT-BAYES and ADA-BARRONS currently constituting its state-of-the-art Pareto frontier. In this paper, we present the first efficient algorithm, BISONS, that obtains polylogarithmic regret with memory and per-step running time requirements that are polynomial in the dimension, displacing ADA-BARRONS from the Pareto frontier. Additionally, we resolve a COLT 2020 open problem by showing that a certain Follow-The-Regularized-Leader algorithm with log-barrier regularization suffers an exponentially larger dependence on the dimension than previously conjectured. Thus, we rule out this algorithm as a candidate for the Pareto frontier. We also extend our algorithm and analysis to a more general problem than online portfolio selection, viz. online learning of quantum states with log loss. This algorithm, called SCHRODINGER'S BISONS, is the first efficient algorithm with polylogarithmic regret for this more general problem.

cs.LG↗

arXiv2022

Megahertz-rate Ultrafast X-ray Scattering and Holographic Imaging at the European XFEL

The advent of X-ray free-electron lasers (XFELs) has revolutionized fundamental science, from atomic to condensed matter physics, from chemistry to biology, giving researchers access to X-rays with unprecedented brightness, coherence, and pulse duration. All XFEL facilities built until recently provided X-ray pulses at a relatively low repetition rate, with limited data statistics. Here, we present the results from the first megahertz repetition rate X-ray scattering experiments at the Spectroscopy and Coherent Scattering (SCS) instrument of the European XFEL. We illustrate the experimental capabilities that the SCS instrument offers, resulting from the operation at MHz repetition rates and the availability of the novel DSSC 2D imaging detector. Time-resolved magnetic X-ray scattering and holographic imaging experiments in solid state samples were chosen as representative, providing an ideal test-bed for operation at megahertz rates. Our results are relevant and applicable to any other non-destructive XFEL experiments in the soft X-ray range.

cond-mat.mes-hall↗

arXiv2022

Machine Learning for Mechanical Ventilation Control

We consider the problem of controlling an invasive mechanical ventilator for pressure-controlled ventilation: a controller must let air in and out of a sedated patient's lungs according to a trajectory of airway pressures specified by a clinician. Hand-tuned PID controllers and similar variants have comprised the industry standard for decades, yet can behave poorly by over- or under-shooting their target or oscillating rapidly. We consider a data-driven machine learning approach: First, we train a simulator based on data we collect from an artificial lung. Then, we train deep neural network controllers on these simulators.We show that our controllers are able to track target pressure waveforms significantly better than PID controllers. We further show that a learned controller generalizes across lungs with varying characteristics much more readily than PID controllers do.

cs.LG↗

arXiv2021

Machine Learning for Mechanical Ventilation Control (Extended Abstract)

Mechanical ventilation is one of the most widely used therapies in the ICU. However, despite broad application from anaesthesia to COVID-related life support, many injurious challenges remain. We frame these as a control problem: ventilators must let air in and out of the patient's lungs according to a prescribed trajectory of airway pressure. Industry-standard controllers, based on the PID method, are neither optimal nor robust. Our data-driven approach learns to control an invasive ventilator by training on a simulator itself trained on data collected from the ventilator. This method outperforms popular reinforcement learning algorithms and even controls the physical ventilator more accurately and robustly than PID. These results underscore how effective data-driven methodologies can be for invasive ventilation and suggest that more general forms of ventilation (e.g., non-invasive, adaptive) may also be amenable.

cs.LG↗

arXiv2021

The Skellam Mechanism for Differentially Private Federated Learning

We introduce the multi-dimensional Skellam mechanism, a discrete differential privacy mechanism based on the difference of two independent Poisson random variables. To quantify its privacy guarantees, we analyze the privacy loss distribution via a numerical evaluation and provide a sharp bound on the Rényi divergence between two shifted Skellam distributions. While useful in both centralized and distributed privacy applications, we investigate how it can be applied in the context of federated learning with secure aggregation under communication constraints. Our theoretical findings and extensive experimental evaluations demonstrate that the Skellam mechanism provides the same privacy-accuracy trade-offs as the continuous Gaussian mechanism, even when the precision is low. More importantly, Skellam is closed under summation and sampling from it only requires sampling from a Poisson distribution -- an efficient routine that ships with all machine learning and data analysis software packages. These features, along with its discrete nature and competitive privacy-accuracy trade-offs, make it an attractive practical alternative to the newly introduced discrete Gaussian mechanism.

cs.LG↗

arXiv2021

Online Target Q-learning with Reverse Experience Replay: Efficiently finding the Optimal Policy for Linear MDPs

Q-learning is a popular Reinforcement Learning (RL) algorithm which is widely used in practice with function approximation (Mnih et al., 2015). In contrast, existing theoretical results are pessimistic about Q-learning. For example, (Baird, 1995) shows that Q-learning does not converge even with linear function approximation for linear MDPs. Furthermore, even for tabular MDPs with synchronous updates, Q-learning was shown to have sub-optimal sample complexity (Li et al., 2021;Azar et al., 2013). The goal of this work is to bridge the gap between practical success of Q-learning and the relatively pessimistic theoretical results. The starting point of our work is the observation that in practice, Q-learning is used with two important modifications: (i) training with two networks, called online network and target network simultaneously (online target learning, or OTL) , and (ii) experience replay (ER) (Mnih et al., 2015). While they have been observed to play a significant role in the practical success of Q-learning, a thorough theoretical understanding of how these two modifications improve the convergence behavior of Q-learning has been missing in literature. By carefully combining Q-learning with OTL and reverse experience replay (RER) (a form of experience replay), we present novel methods Q-Rex and Q-RexDaRe (Q-Rex + data reuse). We show that Q-Rex efficiently finds the optimal policy for linear MDPs (or more generally for MDPs with zero inherent Bellman error with linear approximation (ZIBEL)) and provide non-asymptotic bounds on sample complexity -- the first such result for a Q-learning method for this class of MDPs under standard assumptions. Furthermore, we demonstrate that Q-RexDaRe in fact achieves near optimal sample complexity in the tabular setting, improving upon the existing results for vanilla Q-learning.

cs.LG↗

arXiv2021

Efficient Methods for Online Multiclass Logistic Regression

Multiclass logistic regression is a fundamental task in machine learning with applications in classification and boosting. Previous work (Foster et al., 2018) has highlighted the importance of improper predictors for achieving "fast rates" in the online multiclass logistic regression problem without suffering exponentially from secondary problem parameters, such as the norm of the predictors in the comparison class. While Foster et al. (2018) introduced a statistically optimal algorithm, it is in practice computationally intractable due to its run-time complexity being a large polynomial in the time horizon and dimension of input feature vectors. In this paper, we develop a new algorithm, FOLKLORE, for the problem which runs significantly faster than the algorithm of Foster et al.(2018) -- the running time per iteration scales quadratically in the dimension -- at the cost of a linear dependence on the norm of the predictors in the regret bound. This yields the first practical algorithm for online multiclass logistic regression, resolving an open problem of Foster et al.(2018). Furthermore, we show that our algorithm can be applied to online bandit multiclass prediction and online multiclass boosting, yielding more practical algorithms for both problems compared to the ones in Foster et al.(2018) with similar performance guarantees. Finally, we also provide an online-to-batch conversion result for our algorithm.

cs.LG↗