The rapidly evolving landscape of products, surfaces, policies, and regulations poses significant challenges for deploying state-of-the-art recommendation models at industry scale, primarily due to data fragmentation across domains and escalating infrastructure costs that hinder sustained quality improvements.

Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation—compromising serving quality for efficiency. To address this challenge, we present SOLARIS (Speculative Offloading of Latent-bAsed Representation for Inference Scaling), a novel framework inspired by speculative decoding. SOLARIS proactively precomputes user-item interaction embeddings by predicting which user-item pairs are likely to appear in future requests, and asynchronously generating their foundation model representations ahead of time. This approach decouples the costly foundation model inference from the latency-critical serving path, enabling real-time knowledge transfer from models previously considered too expensive for online use. Deployed across Meta's advertising system serving billions of daily requests, SOLARIS achieves 0.67% revenue-driving top-line metrics gain, demonstrating its effectiveness at scale.

Huge embedding tables in modern deep learning recommender models (DLRM) require prohibitively large memory during training and inference. This paper proposes FIITED, a system to automatically reduce the memory footprint via FIne-grained In-Training Embedding Dimension pruning. By leveraging the key insight that embedding vectors are not equally important, FIITED adaptively adjusts the dimension of each individual embedding vector during model training, assigning larger dimensions to more important embeddings while adapting to dynamic changes in data. We prioritize embedding dimensions with higher frequencies and gradients as more important. To enable efficient pruning of embeddings and their dimensions during model training, we propose an embedding storage system based on virtually-hashed physically-indexed hash tables. Experiments on two industry models and months of realistic datasets show that FIITED can reduce DLRM embedding size by more than 65’ while preserving model quality, outperforming state-of-the-art in-training embedding pruning methods and increasing the reduction ratio by 1.3× to 1.67×. On public datasets, FIITED can reduce the size of embedding tables by 2.2× to 800× with negligible accuracy drop, achieving 1.05× to 12.5× improvement in reduction ratio compared to baselines while improving model throughput.

We study online convex optimization with constraints consisting of multiple functional constraints and a relatively simple constraint set, such as a Euclidean ball. As enforcing the constraints at each time step through projections is computationally challenging in general, we allow decisions to violate the functional constraints but aim to achieve a low regret and cumulative violation of the constraints over a horizon of T time steps. First-order methods achieve an O(sqrt{T}) regret and an O(1) constraint violation, which is the best-known bound under the Slater's condition, but do not take into account the structural information of the problem. Furthermore, the existing algorithms and analysis are limited to Euclidean space. In this paper, we provide an instance-dependent bound for online convex optimization with complex constraints obtained by a novel online primal-dual mirror-prox algorithm. Our instance-dependent regret is quantified by the total gradient variation V_*(T) in the sequence of loss functions. The proposed algorithm works in general normed spaces and simultaneously achieves an O(sqrt{V_*(T)}) regret and an O(1) constraint violation, which is never worse than the best-known (O(sqrt{T}), O(1)) result and improves over previous works that applied mirror-prox-type algorithms for this problem achieving O(T^{2/3}) regret and constraint violation. Finally, our algorithm is computationally efficient, as it only performs mirror descent steps in each iteration instead of solving a general Lagrangian minimization problem.

We study the safe reinforcement learning problem using the constrained Markov decision processes in which an agent aims to maximize the expected total reward subject to a safety constraint on the expected total value of a utility function. We focus on an episodic setting with the function approximation where the Markov transition kernels have a linear structure but do not impose any additional assumptions on the sampling model. Designing safe reinforcement learning algorithms with provable computational and statistical efficiency is particularly challenging under this setting because of the need to incorporate both the safety constraint and the function approximation into the fundamental exploitation/exploration tradeoff. To this end, we present an \underline{O}ptimistic \underline{P}rimal-\underline{D}ual Proximal Policy \underline{OP}timization \mbox{(OPDOP)} algorithm where the value function is estimated by combining the least-squares policy evaluation and an additional bonus term for safe exploration. We prove that the proposed algorithm achieves an $\tilde{O}(d H^{2.5}\sqrt{T})$ regret and an $\tilde{O}(d H^{2.5}\sqrt{T})$ constraint violation, where $d$ is the dimension of the feature mapping, $H$ is the horizon of each episode, and $T$ is the total number of steps. These bounds hold when the reward/utility functions are fixed but the feedback after each episode is bandit. Our bounds depend on the capacity of the state-action space only through the dimension of the feature mapping and thus our results hold even when the number of states goes to infinity. To the best of our knowledge, we provide the first provably efficient online policy optimization algorithm for constrained Markov decision processes in the function approximation setting, with safe exploration.

While single-agent policy optimization in a fixed environment has attracted a lot of research attention recently in the reinforcement learning community, much less is known theoretically when there are multiple agents playing in a potentially competitive environment. We take steps forward by proposing and analyzing new fictitious play policy optimization algorithms for two-player zero-sum Markov games with structured but unknown transitions. We consider two classes of transition structures: factored independent transition and single-controller transition. For both scenarios, we prove tight $\widetilde{\mathcal{O}}(\sqrt{T})$ regret bounds after $T$ steps in a two-agent competitive game scenario. The regret of each player is measured against a potentially adversarial opponent who can choose a single best policy in hindsight after observing the full policy sequence. Our algorithms feature a combination of Upper Confidence Bound (UCB)-type optimism and fictitious play under the scope of simultaneous policy optimization in a non-stationary environment. When both players adopt the proposed algorithms, their overall optimality gap is $\widetilde{\mathcal{O}}(\sqrt{T})$.

In this paper, we consider hybrid parallelism---a paradigm that employs both Data Parallelism (DP) and Model Parallelism (MP)---to scale distributed training of large recommendation models. We propose a compression framework called Dynamic Communication Thresholding (DCT) for communication-efficient hybrid training. DCT filters the entities to be communicated across the network through a simple hard-thresholding function, allowing only the most relevant information to pass through. For communication efficient DP, DCT compresses the parameter gradients sent to the parameter server during model synchronization. The threshold is updated only once every few thousand iterations to reduce the computational overhead of compression. For communication efficient MP, DCT incorporates a novel technique to compress the activations and gradients sent across the network during the forward and backward propagation, respectively. This is done by identifying and updating only the most relevant neurons of the neural network for each training sample in the data. We evaluate DCT on publicly available natural language processing and recommender models and datasets, as well as recommendation systems used in production at Facebook. DCT reduces communication by at least 100x and 20x during DP and MP, respectively. The algorithm has been deployed in production, and it improves end-to-end training time for a state-of-the-art industrial recommender model by 37%, without any loss in performance.

Neural network based recommendation models are widely used to power many internet-scale applications including product recommendation and feed ranking. As the models become more complex and more training data is required during training, improving the training scalability of these recommendation models becomes an urgent need. However, improving the scalability without sacrificing the model quality is challenging. In this paper, we conduct an in-depth analysis of the scalability bottleneck in existing training architecture on large scale CPU clusters. Based on these observations, we propose a new training architecture called Hierarchical Training, which exploits both data parallelism and model parallelism for the neural network part of the model within a group. We implement hierarchical training with a two-layer design: a tagging system that decides the operator placement and a net transformation system that materializes the training plans, and integrate hierarchical training into existing training stack. We propose several optimizations to improve the scalability of hierarchical training including model architecture optimization, communication compression, and various system-level improvements. Extensive experiments at massive scale demonstrate that hierarchical training can speed up distributed recommendation model training by 1.9x without model quality drop.

We study the robust one-bit compressed sensing problem whose goal is to design an algorithm that faithfully recovers any sparse target vector $\theta_0\in\mathbb{R}^d$ \emph{uniformly} via $m$ quantized noisy measurements. Specifically, we consider a new framework for this problem where the sparsity is implicitly enforced via mapping a low dimensional representation $x_0 \in \mathbb{R}^k$ through a known $n$-layer ReLU generative network $G:\mathbb{R}^k\rightarrow\mathbb{R}^d$ such that $\theta_0 = G(x_0)$. Such a framework poses low-dimensional priors on $\theta_0$ without a known sparsity basis. We propose to recover the target $G(x_0)$ solving an unconstrained empirical risk minimization (ERM). Under a weak \emph{sub-exponential measurement assumption}, we establish a joint statistical and computational analysis. In particular, we prove that the ERM estimator in this new framework achieves a statistical rate of $m=\widetilde{\mathcal{O}}(kn \log d /\varepsilon^2)$ recovering any $G(x_0)$ uniformly up to an error $\varepsilon$. When the network is shallow (i.e., $n$ is small), we show this rate matches the information-theoretic lower bound up to logarithm factors of $\varepsilon^{-1}$. From the lens of computation, we prove that under proper conditions on the network weights, our proposed empirical risk, despite non-convexity, has no stationary point outside of small neighborhoods around the true representation $x_0$ and its negative multiple; furthermore, we show that the global minimizer of the empirical risk stays within the neighborhood around $x_0$ rather than its negative multiple under further assumptions on weights.

We consider online learning for episodic stochastically constrained Markov decision processes (CMDP), which plays a central role in ensuring the safety of reinforcement learning. Here the loss function can vary arbitrarily across the episodes, whereas both the loss received and the budget consumption are revealed at the end of each episode. Previous works solve this problem under the restrictive assumption that the transition model of the MDP is known a priori and establish regret bounds that depend polynomially on the cardinalities of the state space $\mathcal{S}$ and the action space $\mathcal{A}$. In this work, we propose a new \emph{upper confidence primal-dual} algorithm, which only requires the trajectories sampled from the transition model. In particular, we prove that the proposed algorithm achieves $\widetilde{\mathcal{O}}(L|\mathcal{S}|\sqrt{|\mathcal{A}|T})$ upper bounds of both the regret and the constraint violation, where $L$ is the length of each episode. Our analysis incorporates a new high-probability drift analysis of Lagrange multiplier processes into the celebrated regret analysis of upper confidence reinforcement learning, which demonstrates the power of ``optimism in the face of uncertainty'' in constrained online learning.

We consider estimating a high-dimensional vector from non-linear measurements where the unknown vector is represented by a generative model $G:\mathbb{R}^k\rightarrow\mathbb{R}^d$ with $k\ll d$. Such a model poses structural priors on the unknown vector without having a dedicated basis, and in particular allows new and efficient approaches solving recovery problems with number of measurements far less than the ambient dimension of the vector. While progresses have been made recently regarding theoretical understandings on the linear Gaussian measurements, much less is known when the model is possibly misspecified and the measurements are non-Gaussian. In this paper, we make a step towards such a direction by considering the scenario where the measurements are non-Gaussian, subject to possibly unknown nonlinear transformations and the responses are heavy-tailed. We then propose new estimators via score functions based on the first and second order Stein’s identity, and prove the sample size bound of $m=\mathcal{O}(k\varepsilon^{-2}\log(L/\varepsilon))$ achieving an $\varepsilon$ error in the form of exponential concentration inequalities. Furthermore, for the special case of multi-layer ReLU generative model, we improve the sample bound by a logarithm factor to $m=\mathcal{O}(k\varepsilon^{-2}\log(d))$, matching the state-of-art statistical rate in compressed sensing for estimating $k$-sparse vectors. On the technical side, we develop new chaining methods bounding heavy-tailed processes, which could be of independent interest.

We propose a new primal-dual homotopy smoothing algorithm for a linearly constrained convex program, where neither the primal nor the dual function has to be smooth or strongly convex. The best known iteration complexity solving such a non-smooth problem is $\mathcal{O}(\varepsilon^{-1})$. In this paper, we show that by leveraging a local error bound condition on the dual function, the proposed algorithm can achieve a better primal convergence time of $\mathcal{O}\l(\varepsilon^{-2/(2+\beta)}\log_2(\varepsilon^{-1})\r)$, where $\beta\in(0,1]$ is a local error bound parameter. As an example application, we show that the distributed geometric median problem, which can be formulated as a constrained convex program, has its dual function non-smooth but satisfying the aforementioned local error bound condition with $\beta=1/2$, therefore enjoying a convergence time of $\mathcal{O}\l(\varepsilon^{-4/5}\log_2(\varepsilon^{-1})\r)$. This result improves upon the $\mathcal{O}(\varepsilon^{-1})$ convergence time bound achieved by existing distributed optimization algorithms. Simulation experiments also demonstrate the performance of our proposed algorithm.

We study high-dimensional signal recovery from non-linear measurements with design vectors having elliptically symmetric distribution. Special attention is devoted to the situation when the unknown signal belongs to a set of low statistical complexity, while both the measurements and the design vectors are heavy-tailed. We propose and analyze a new estimator that adapts to the structure of the problem, while being robust both to the possible model misspecification characterized by arbitrary non-linearity of the measurements as well as to data corruption modeled by the heavy-tailed distributions. Moreover, this estimator has low computational complexity. Our results are expressed in the form of exponential concentration inequalities for the error of the proposed estimator. On the technical side, our proofs rely on the generic chaining methods, and illustrate the power of this approach for statistical applications. Theory is supported by numerical experiments demonstrating that our estimator outperforms existing alternatives when data are heavy-tailed.

We propose and analyze a new estimator of the covariance matrix that admits strong theoretical guarantees under weak assumptions on the underlying distribution, such as existence of moments of only low order. While estimation of covariance matrices corresponding to sub-Gaussian distributions is well-understood, much less in known in the case of heavy-tailed data. As K. Balasubramanian and M. Yuan write, data from real-world experiments oftentimes tend to be corrupted with outliers and/or exhibit heavy tails. In such cases, it is not clear that those covariance matrix estimators .. remain optimal'' and..what are the other possible strategies to deal with heavy tailed distributions warrant further studies.'' We make a step towards answering this question and prove tight deviation inequalities for the proposed estimator that depend only on the parameters controlling the ``intrinsic dimension'' associated to the covariance matrix (as opposed to the dimension of the ambient space); in particular, our results are applicable in the case of high-dimensional observations.

This paper considers online convex optimization (OCO) with stochastic constraints, which generalizes Zinkevich's OCO over a known simple fixed set by introducing multiple stochastic functional constraints that are i.i.d. generated at each round and are disclosed to the decision maker only after the decision is made. This formulation arises naturally when decisions are restricted by stochastic environments or deterministic environments with noisy observations. It also includes many important problems as special case, such as OCO with long term constraints, stochastic constrained convex optimization, and deterministic constrained convex optimization. To solve this problem, this paper proposes a new algorithm that achieves $O(\sqrt{T})$ expected regret and constraint violations and $O(\sqrt{T}\log(T))$ high probability regret and constraint violations. Experiments on a real-world data center scheduling problem further verify the performance of the new algorithm.

This paper considers a cost minimization problem for data centers with N servers and randomly arriving service requests. A central router decides which server to use for each new request. Each server has three types of states (active, idle, and setup) with different costs and time durations. The servers operate asynchronously over their own states and can choose one of multiple sleep modes when idle. We develop an online distributed control algorithm so that each server makes its own decisions. The request queues are bounded and the overall time average cost is near optimal with probability 1. First the algorithm does not need probability information for the arrival rate or job sizes. Finally, an improved algorithm that uses a single queue is developed via a “virtualization” technique, which is shown to provide the same (near optimal) costs. Simulation experiments on a real data center traffic trace demonstrate the efficiency of our algorithm compared with other existing algorithms.

This paper considers optimization of power and delay in a time-varying wireless link using rateless codes. The link serves a sequence of variable-length packets. Each packet is coded and transmitted over multiple slots. Channel conditions can change from slot to slot and are unknown to the transmitter. The amount of mutual information accumulated on each slot depends on the random channel realization and the power used. The goal is to minimize average service delay subject to an average power constraint. We formulate this problem as a frame-based stochastic optimization problem and solve it via an online algorithm. We show that the subproblem within each frame is a simple integer program which can be effectively solved using a dynamic program. The optimality of this online algorithm is proved using the frame-based Lyapunov drift analysis.

This paper treats power-aware throughput maximization in a multiuser file downloading system. Each user can receive a new file only after its previous file is finished. The file state processes for each user act as coupled Markov chains that form a generalized restless bandit system. First, an optimal algorithm is derived for the case of one user. The algorithm maximizes throughput subject to an average power constraint. Next, the one-user algorithm is extended to a low-complexity heuristic for the multiuser problem. The heuristic uses a simple online index policy. In a special case with no power-constraint, the multiuser heuristic is shown to be throughput-optimal. Simulations are used to demonstrate effectiveness of the heuristic in the general case. For simple cases where the optimal solution can be computed offline, the heuristic is shown to be near-optimal for a wide range of parameters.

This paper treats power-aware throughput maximization in a multi-user file downloading system. Each user can receive a new file only after its previous file is finished. The file state processes for each user act as coupled Markov chains that form a generalized restless bandit system. First, an optimal algorithm is derived for the case of one user. The algorithm maximizes throughput subject to an average power constraint. Next, the one-user algorithm is extended to a low complexity heuristic for the multi-user problem. The heuristic uses a simple online index policy and its effectiveness is shown via simulation. For simple 3-user cases where the optimal solution can be computed offline, the heuristic is shown to be near-optimal for a wide range of parameters.

In this letter, a new DOA estimation method based on sparse signal recovery is proposed. We utilize the Capon spectrum to design a weightedl1-norm penalty in order to further enforce the sparsity and approximate the originall0-norm. A theoretical guidance for choosing a proper regularization parameter is also presented according to the dual form of the original problem. Simulation results demonstrate the effectiveness and efficiency of the proposed method.