The emergence of Large Language Models (LLMs) with strong reasoning capabilities marks a significant milestone, unlocking new frontiers in complex problem-solving. However, training these reasoning models, typically using Reinforcement Learning (RL), encounters critical efficiency bottlenecks: response generation during RL training exhibits a persistent long-tail distribution, where a few very long responses dominate execution time, wasting resources and inflating costs. To address this, we propose TLT, a system that accelerates reasoning RL training losslessly by integrating adaptive speculative decoding. Applying speculative decoding in RL is challenging due to the dynamic workloads, evolving target model, and draft model training overhead. TLT overcomes these obstacles with two synergistic components: (1) Adaptive Drafter, a lightweight draft model trained continuously on idle GPUs during long-tail generation to maintain alignment with the target model at no extra cost; and (2) Adaptive Rollout Engine, which maintains a memory-efficient pool of pre-captured CUDAGraphs and adaptively select suitable SD strategies for each input batch. Evaluations demonstrate that TLT achieves over 1.7x end-to-end RL training speedup over state-of-the-art systems, preserves the model accuracy, and yields a high-quality draft model as a free byproduct suitable for efficient deployment. Code is released at https://github.com/mit-han-lab/fastrl.

Synchronization errors, including insertions, deletions, and substitutions, may occur in bursts in communication systems such as DNA data storage, file synchronization, and magnetic recording. In this paper, we study an error model con?sisting of one burst of t-deletions and one burst of t-substitutions. By reformulating the original sequence into a matrix form, we propose an explicit construction of error-correcting codes capable of correcting one burst of t-deletions and one burst of t-substitutions with O(log n) redundancy.

In this paper, we consider convertible codes with the locally repairable property. We present an improved lower bound on access cost associated with (r, δ ⩾ 2)-locality, which extends the known lower bound only related tor.We then provide a general construction of convertible codes with optimal access cost which shows that these codes can have super-linear length or maximally recoverable property. Furthermore, we propose explicit constructions of convertible codes with the final code achieving super-linear length or maximally recoverable property when the initial codes have super-linear length or maximally recoverable property respectively. Specifically, compared with known constructions, our construction is applicable to the case δ > 2 for the first time.

High-resolution perception of visual details is crucial for daily tasks. Current vision pre-training, however, is still limited to low resolutions (e.g., 378×378 pixels) due to the quadratic cost of processing larger images. We introduce PS3 that scales CLIP-style vision pre-training to 4K resolution with a near-constant cost. Instead of contrastive learning on global image representation, PS3 is pre-trained by selectively processing local regions and contrasting them with local detailed captions, enabling high-resolution representation learning with greatly reduced computational overhead. The pre-trained PS3 is able to both encode the global image at low resolution and selectively process local high-resolution regions based on their saliency or relevance to a text prompt. When applying PS3 to multi-modal LLM (MLLM), the resulting model, named VILA-HD, significantly improves high-resolution visual perception compared to baselines without high-resolution vision pre-training such as AnyRes and S2while using up to 4.3× fewer tokens. PS3 also unlocks appealing scaling properties of VILA-HD, including scaling up resolution for free and scaling up test-time compute for better performance. Compared to state of the arts, VILA-HD outperforms previous MLLMs such as NVILA and Qwen2-VL across multiple benchmarks and achieves better efficiency than latest token pruning approaches. Finally, we find current benchmarks do not require 4K-resolution perception, which motivates us to propose 4KPro, a new benchmark of image QA at 4K resolution, on which VILA-HD outperforms all previous MLLMs, including a 14.5% improvement over GPT-4o, and a 3.2% improvement and 2.96× speedup over Qwen2-VL.

This paper presents SANA-Sprint, an efficient diffusion model for ultra-fast text-to-image (T2I) generation. SANA-Sprint is built on a pre-trained foundation model and augmented with hybrid distillation, dramatically reducing inference steps from 20 to 1-4. We introduce three key innovations: (1) We propose a training-free approach that transforms a pre-trained flow-matching model for continuous-time consistency distillation (sCM), eliminating costly training from scratch and achieving high training efficiency. Our hybrid distillation strategy combines sCM with latent adversarial distillation (LADD): sCM ensures alignment with the teacher model, while LADD enhances single-step generation fidelity. (2) SANA-Sprint is a unified step-adaptive model that achieves high-quality generation in 1-4 steps, eliminating step-specific training and improving efficiency. (3) We integrate ControlNet with SANA-Sprint for real-time interactive image generation, enabling instant visual feedback for user interaction. SANA-Sprint establishes a new Pareto frontier in speed-quality tradeoffs, achieving state-of-the-art performance with 7.59 FID and 0.74 GenEval in only 1 step - outperforming FLUX-schnell (7.94 FID / 0.71 GenEval) while being 10x faster (0.1s vs 1.1s on H100). It also achieves 0.1s (T2I) and 0.25s (ControlNet) latency for 1024 x 1024 images on H100, and 0.31s (T2I) on an RTX 4090, showcasing its exceptional efficiency and potential for AI-powered consumer applications (AIPC). Code and pre-trained models will be open-sourced.

We present DC-AE 1.5, a new family of deep compression autoencoders for high-resolution diffusion models. Increasing the autoencoder's latent channel number is a highly effective approach for improving its reconstruction quality. However, it results in slow convergence for diffusion models, leading to poorer generation quality despite better reconstruction quality. This issue limits the quality upper bound of latent diffusion models and hinders the employment of autoencoders with higher spatial compression ratios. We introduce two key innovations to address this challenge: i) Structured Latent Space, a training-based approach to impose a desired channel-wise structure on the latent space with front latent channels capturing object structures and latter latent channels capturing image details; ii) Augmented Diffusion Training, an augmented diffusion training strategy with additional diffusion training objectives on object latent channels to accelerate convergence. With these techniques, DC-AE 1.5 delivers faster convergence and better diffusion scaling results than DC-AE. On ImageNet 512x512, DC-AE-1.5-f64c128 delivers better image generation quality than DC-AE-f32c32 while being 4x faster. Code: https://github.com/dc-ai-projects/DC-Gen.

We introduce DC-AR, a novel masked autoregressive (AR) text-to-image generation framework that delivers superior image generation quality with exceptional computational efficiency. Due to the tokenizers' limitations, prior masked AR models have lagged behind diffusion models in terms of quality or efficiency. We overcome this limitation by introducing DC-HT - a deep compression hybrid tokenizer for AR models that achieves a 32x spatial compression ratio while maintaining high reconstruction fidelity and cross-resolution generalization ability. Building upon DC-HT, we extend MaskGIT and create a new hybrid masked autoregressive image generation framework that first produces the structural elements through discrete tokens and then applies refinements via residual tokens. DC-AR achieves state-of-the-art results with a gFID of 5.49 on MJHQ-30K and an overall score of 0.69 on GenEval, while offering 1.5-7.9x higher throughput and 2.0-3.5x lower latency compared to prior leading diffusion and autoregressive models.

We present Deep Compression Autoencoder (DC-AE), a new family of autoencoders for accelerating high-resolution diffusion models. Existing autoencodes have demonstrated impressive results at a moderate spatial compression ratio (e.g., 8x), but fail to maintain satisfactory reconstruction accuracy for high spatial compression ratios (e.g., 64x). We address this challenge by introducing two key techniques: (1) Residual Autoencoding, where we design our models to learn residuals based on the space-to-channel transformed features to alleviate the optimization difficulty of high spatial-compression autoencoders; (2) Decoupled High-Resolution Adaptation, an efficient decoupled three-phase training strategy for mitigating the generalization penalty of high spatial-compression autoencoders. With these designs, we improve the autoencoder's spatial compression ratio up to 128 while maintaining the reconstruction quality. Applying our DC-AE to latent diffusion models, we achieve significant speedup without accuracy drop. For example, on ImageNet 512x512, our DC-AE provides 19.1x inference speedup and 17.9x training speedup on H100 GPU for UViT-H while achieving a better FID, compared with the widely used SD-VAE-f8 autoencoder.

We introduce Hybrid Autoregressive Transformer (HART), the first autoregressive (AR) visual generation model capable of directly generating 1024x1024 images, rivaling diffusion models in image generation quality. Existing AR models face limitations due to the poor image reconstruction quality of their discrete tokenizers and the prohibitive training costs associated with generating 1024px images. To address these challenges, we present the hybrid tokenizer, which decomposes the continuous latents from the autoencoder into two components: discrete tokens representing the big picture and continuous tokens representing the residual components that cannot be represented by the discrete tokens. The discrete component is modeled by a scalable-resolution discrete AR model, while the continuous component is learned with a lightweight residual diffusion module with only 37M parameters. Compared with the discrete-only VAR tokenizer, our hybrid approach improves reconstruction FID from 2.11 to 0.30 on MJHQ-30K, leading to a 31% generation FID improvement from 7.85 to 5.38. HART also outperforms state-of-the-art diffusion models in both FID and CLIP score, with 4.5-7.7$\times$ higher throughput and 6.9-13.4$\times$ lower MACs. Our code is open sourced at https://github.com/mit-han-lab/hart.

FP8 training has emerged as a promising method for improving training efficiency. Existing frameworks accelerate training by applying FP8 computation to linear layers while leaving optimizer states and activations in higher precision, which fails to fully optimize memory usage. This paper introduces COAT (**C**ompressing **O**ptimizer States and **A**ctivations for FP8 **T**raining), a novel FP8 training framework designed to significantly reduce memory footprint when training large models. COAT addresses current limitations through two key innovations: (1) **Dynamic Range Expansion**, which aligns optimizer state distributions more closely with the FP8 representation range, thereby reducing quantization error, and (2) **Mixed-Granularity Activation Quantization**, which optimizes activation memory using a combination of per-tensor and per-group quantization strategies. Experiments demonstrate that COAT effectively reduces end-to-end training memory footprint by **1.54×** compared to BF16 while achieving nearly lossless performance across various tasks, such as Large Language Model pretraining and fine-tuning and Vision Language Model training. COAT also achieves a **1.43×** end-to-end training speedup compared to BF16, performing on par with or surpassing TransformerEngine's speedup. COAT enables efficient full-parameter training of large models on fewer GPUs, and facilitates doubling the batch size in distributed training settings, providing a practical solution for scaling large-scale model training. Code will be released upon publication.

We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096$\times$4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8$\times$, we trained an AE that can compress images 32$\times$, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolutions without sacrificing quality. (3) Decoder-only text encoder: we replaced T5 with modern decoder-only small LLM as the text encoder and designed complex human instruction with in-context learning to enhance the image-text alignment. (4) Efficient training and sampling: we propose Flow-DPM-Solver to reduce sampling steps, with efficient caption labeling and selection to accelerate convergence. As a result, Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput. Moreover, Sana-0.6B can be deployed on a 16GB laptop GPU, taking less than 1 second to generate a 1024$\times$1024 resolution image. Sana enables content creation at low cost. Code and model will be publicly released upon publication.

Diffusion Transformers (DiTs) dominate video generation but their high computational cost severely limits real-world applicability, usually requiring tens of minutes to generate a few seconds of video even on high-performance GPUs. This inefficiency primarily arises from the quadratic computational complexity of 3D full attention with respect to the context length. In this paper, we propose a training-free framework termed Sparse VideoGen (SVG) that leverages the inherent sparsity in 3D full attention to boost inference efficiency. We reveal that the attention heads can be dynamically classified into two groups depending on distinct sparse patterns: (1) Spatial Head, where only spatially-related tokens within each frame dominate the attention output, and (2) Temporal Head, where only temporally-related tokens across different frames dominate. Based on this insight, SVG proposes an online profiling strategy to capture the dynamic sparse patterns and predicts the type of attention head. Combined with a novel hardware-efficient tensor layout transformation and customized kernel implementations, SVG achieves up to 2.28$\times$ and 2.33$\times$ end-to-end speedup on CogVideoX-v1.5 and HunyuanVideo, respectively, while preserving generation quality. Our code will be open-sourced upon publication.

This paper presents SANA-1.5, a linear Diffusion Transformer for efficient scaling in text-to-image generation. Building upon SANA-1.0, we introduce three key innovations: (1) Efficient Training Scaling: A depth-growth paradigm that enables scaling from 1.6B to 4.8B parameters with significantly reduced computational resources, combined with a memory-efficient 8-bit optimizer. (2) Model Depth Pruning: A block importance analysis technique for efficient model compression to arbitrary sizes with minimal quality loss. (3) Inference-time Scaling: A repeated sampling strategy that trades computation for model capacity, enabling smaller models to match larger model quality at inference time. Through these strategies, SANA-1.5 achieves a text-image alignment score of 0.72 on GenEval, which can be further improved to 0.80 through inference scaling, establishing a new SoTA on GenEval benchmark. These innovations enable efficient model scaling across different compute budgets while maintaining high quality, making high-quality image generation more accessible.

In this paper, we address the multi-node recovery problem for Reed-Solomon (RS) codes. To overcome the obstacle that most existing cooperative schemes can only deal with scenarios involving fewer than three failures, we consider a new repair model, referred to as the individual repair mode, in which each failed node can be repaired individually. Building upon this model, we propose a repair scheme for$[n, k]$Reed-Solomon codes that is applicable to any$0

We present Jet-Nemotron, a new family of hybrid-architecture language models, which matches or exceeds the accuracy of leading full-attention models while significantly improving generation throughput. Jet-Nemotron is developed using Post Neural Architecture Search (PostNAS), a novel neural architecture exploration pipeline that enables efficient model design. Unlike prior approaches, PostNAS begins with a pre-trained full-attention model and freezes its MLP weights, allowing efficient exploration of attention block designs. The pipeline includes four key components: (1) learning optimal full-attention layer placement and elimination, (2) linear attention block selection, (3) designing new attention blocks, and (4) performing hardware-aware hyperparameter search. Our Jet-Nemotron-2B model achieves comparable or superior accuracy to Qwen3, Qwen2.5, Gemma3, and Llama3.2 across a comprehensive suite of benchmarks while delivering up to 53.6× generation throughput speedup and 6.1× prefilling speedup. It also achieves higher accuracy on MMLU and MMLU-Pro than recent advanced MoE full-attention models, such as DeepSeek-V3-Small and Moonlight, despite their larger scale with 15B total and 2.2B activated parameters.

Large language models (LLMs) have shown remarkable performance across diverse reasoning and generation tasks, and are increasingly deployed as agents in dynamic environments such as code generation and recommendation systems. However, many real-world applications, such as high-frequency trading and real-time competitive gaming, require decisions under strict latency constraints, where faster responses directly translate into higher rewards. Despite the importance of this latency–quality trade-off, it remains underexplored in the context of LLM-based agents. In this work, we present the first systematic study of this trade-off in real-time decision-making tasks. To support our investigation, we introduce two new benchmarks: HFTBench, a high-frequency trading simulation, and StreetFighter, a competitive gaming platform. Our analysis reveals that optimal latency–quality balance varies by task, and that sacrificing quality for lower latency can significantly enhance downstream performance. To address this, we propose FPX, an adaptive framework that dynamically selects model size and quantization level based on real-time demands. Our method achieves the best performance on both benchmarks, improving win rate by up to 80% in Street Fighter and boosting daily yield by up to 26.52% in trading, underscoring the need for latency-aware evaluation and deployment strategies for LLM-based agents. These results demonstrate the critical importance of latency-aware evaluation and deployment strategies for real-world LLM-based agents.

Diffusion Transformers (DiTs) are essential for video generation but suffer from significant latency due to the quadratic complexity of attention. By computing only critical tokens, sparse attention reduces computational costs and offers a promising acceleration approach. However, we identify that existing methods fail to approach optimal generation quality under the same computation budget for two reasons: (1) Inaccurate critical token identification: current methods cluster tokens based on position rather than semantics, leading to imprecise aggregated representations. (2) Excessive computation waste: critical tokens are scattered among non-critical ones, leading to wasted computation on GPUs, which are optimized for processing contiguous tokens. In this paper, we propose SVG2, a training-free framework that maximizes identification accuracy and minimizes computation waste, achieving a Pareto frontier trade-off between generation quality and efficiency. The core of SVG2 is semantic-aware permutation, which clusters and reorders tokens based on semantic similarity using k-means. This approach ensures both a precise cluster representation, improving identification accuracy, and a densified layout of critical tokens, enabling efficient computation without padding. Additionally, SVG2 integrates Top-p dynamic budget control and customized kernel implementations, achieving up to $2.30\times$ and $1.89\times$ speedup while maintaining a PSNR of up to $30$ and $26$ on HunyuanVideo and Wan 2.1, respectively. Our code is open-sourced at https://github.com/svg-project/Sparse-VideoGen.

Linear feedback coding scheme, such as the elegant Schalkwijk-Kailath (SK) scheme, receives much attention in the literature since its decoding error probability decreases as a second-order exponential in the coding blocklength. In recent years, a linear feedback scheme has been proposed for the Gaussian relay channel (GRC) with destination-source feedback, which combines the SK scheme and the amplify-and-forward (AF) relay strategy. Since there exists three possible feedback links in the GRC, then one question beckons: is there any rate gain if there exist multiple feedback links in the GRC, and can any other relay strategy outperform the AF strategy? In this paper, we answer this question by investigating four feedback models of the GRC, namely, the GRC with destination-source and destination-relay feedback, the GRC with destination-relay and relay-source feedback, the GRC with destination-source and relay-source feedback, and the GRC with all feedback links, respectively. We propose SK-type schemes for these feedback models, and numerical examples show that when the coding blocklength is not long, the rates of our proposed schemes almost approach their asymptotic values, and these rates may be larger than those of existing schemes in the literature. The study of this paper shows that different number/location of feedback links may bring additional rate gain in finite blocklength regime.

Maximum distance separable (MDS) codes in distributed storage systems provide the optimal tradeoff between fault tolerance and storage overhead. As a kind of MDS codes, minimum storage regenerating (MSR) codes have attracted a lot of attention since they are also optimal in terms of repair bandwidth. However, MSR codes suffer from a high repair degree, meaning many helper nodes are needed in the node repair process. Compared to MSR codes, locally repairable codes (LRCs) can significantly reduce the repair degree at the cost of increased storage overhead. The recently introduced concept of vector LRCs combines the advantages of MSR codes and LRCs, providing a tradeoff between repair degree/repair bandwidth and storage overhead. Most existing vector LRCs are built on MSR codes or their shortened versions. However, existing MSR codes have an unavoidably large sub-packetization levels, which also result in large sub-packetization levels in the corresponding vector LRCs. In this paper, we propose a new vector LRC structure, where MDS array codes (without shortening) can be employed as the local codes. Based this new structure, we propose three constructions of vector LRCs with small sub-packetization levels and small repair bandwidth, whose required field sizes are comparable to the code lengths. Additionally, the first two constructions offer a flexible tradeoff between the sub-packetization level and the repair bandwidth, while the third construction has a sub-packetization level of 2, making it easy to implement. Compared to existing vector LRCs, the new vector LRCs provide significantly smaller sub-packetization levels and support a wider range of parameters.

In this paper, the repair problem for erasures beyond locality in locally repairable codes is explored under a practical system setting, where a rack-aware storage system consists of racks, each containing a few parity checks. This is referred to as a rack-aware system with locality. Two repair schemes are devised to reduce the repair bandwidth for Tamo-Barg codes under the rack-aware model by setting each repair set as a rack. Additionally, a cut-set bound for locally repairable codes under the rack-aware model with locality is introduced. Using this bound, the second repair scheme is proven to be optimal. Furthermore, the partial-repair problem is considered for locally repairable codes under the rack-aware model with locality, and both repair schemes and bounds are introduced for this scenario.n this paper, the repair problem for erasures beyond locality in locally repairable codes is explored under a practical system setting, where a rack-aware storage system consists of racks, each containing a few parity checks. This is referred to as a rack-aware system with locality. Two repair schemes are devised to reduce the repair bandwidth for Tamo-Barg codes under the rack-aware model by setting each repair set as a rack. Additionally, a cut-set bound for locally repairable codes under the rack-aware model with locality is introduced. Using this bound, the second repair scheme is proven to be optimal. Furthermore, the partial-repair problem is considered for locally repairable codes under the rack-aware model with locality, and both repair schemes and bounds are introduced for this scenario.

We present Condition-Aware Neural N etwork (CAN), a new method for adding control to image generative models. In parallel to prior conditional control methods, CAN controls the image generation process by dynamically manipulating the weight of the neural network. This is achieved by introducing a condition-aware weight generation module that generates conditional weight for convolution/linear layers based on the input condition. We test CAN on class-conditional image generation on ImageNet and text-to-image generation on COCO. CAN consistently delivers significant improvements for diffusion transformer models, including DiT and UViT. In particular, CA N combined with EfficientViT (CaT) achieves 2.78 FID on ImageNet$5l2\times 5l2$, surpassing DiT-XL/2 while requiring$52\times$fewer MACs per sampling step.

Diffusion models have achieved great success in synthesizing high-quality images. However, generating high-resolution images with diffusion models is still challenging due to the enormous computational costs, resulting in a pro-hibitive latency for interactive applications. In this paper, we propose DistriFusion to tackle this problem by leveraging parallelism across multiple GPUs. Our method splits the model input into multiple patches and assigns each patch to a GPU. However,naïvely implementing such an algorithm breaks the interaction between patches and loses fidelity, while incorporating such an interaction will incur tremendous communication overhead. To overcome this dilemma, we observe the high similarity between the input from adjacent diffusion steps and propose displaced patch parallelism, which takes advantage of the sequential nature of the diffusion process by reusing the pre-computed feature maps from the previous timestep to provide context for the current step. Therefore, our method supports asynchronous commu-nication, which can be pipelined by computation. Extensive experiments show that our method can be applied to recent Stable Diffusion XL with no quality degradation and achieve up to a 6.1 ×speedup on eight A100 GPUs compared to one.

Quantifying negative information in an efficient way is a challenging task, especially when this information has to be communicated on a network. In this article we leverage the unique properties offered by possibility theory to quantify and approximate the negative information arising in the context of tracking a target with a sensor that has a limited field of view. We also verify experimentally that the corresponding target tracking methodology can be applied in a decentralised manner to a sensor network, while maintaining a performance close to the idealised case where the initial location of the target is better-known.

Open-pit mining areas have special geological structures, complex road networks, multi-rotation sections and poor road conditions, which bring many challenges to the operation of mining autonomous transport vehicles. Due to the large size and high control difficulty of mining autonomous transport vehicles, poor control effect and inaccurate steering mechanism implementation are prone to occur during the driving process, which leads to the vehicle departure from the reference path. To ensure the safety of autonomous operation in intelligent mine, this paper proposes a method for predicting driving departure of mining autonomous vehicles based on Gated Recurrent Unit (GRU). Firstly, the vehicle's historical trajectory data is obtained via on-board sensors, followed by data cleaning and normalization processes. Key features are extracted, and a vehicle driving scene recognition module based on a GRU-based network is designed using deep learning. Subsequently, a GA-seq2seqGRU trajectory prediction module is constructed utilizing the scene recognition results, and the Genetic Algorithms (GA) is employed to tune the network hyper-parameters. Based on the predicted trajectories, the overall departure risk is calculated using two departure judgment methods based on cross-lane time and predicted lateral deviation, which are mapped to the departure level. Simulation experiments demonstrate that the accuracy of the driving scene recognition model in the proposed method in this paper reaches 0.9721, which is better than the 0.9585 of the Long Short Memory Neural Network (LSTM) model, and the trajectory prediction model based on the results of the driving scene recognition has a smaller RMSE value than the LSTM, GRU, and GA-seq2seqLSTM models when the prediction time domains are 1s, 2s, 3s, 4s, and 5s, and the departure detection module The average time consumed is 0.142ms.

In this study, we tackle the path-tracking problem of a nonlinear four-wheel steering vehicle dynamics model subject to model mismatches and propose a model predictive control (MPC) algorithm based on an event-triggered mechanism (ET -MPC). The goal is to maintain closed-loop control performance while reducing the computational and communication burdens of traditional MPC. We introduce an ET -MPC framework utilizing a model-free reinforcement learning agent with proximal policy optimization (PPO). This agent interacts with the MPC system, progressively learning to determine the optimal event-triggered mechanism. To enhance exploration and training efficiency, we incorporate the Long Short-Term Memory (LSTM) technique into PPO. Experimental results show that the proposed ET -MPC framework, combined with reinforcement learning for reward optimization, demonstrates superior overall performance in path-tracking control of four-wheel steering vehicles.

We study the problem of repairing erasures in locally repairable codes beyond the code locality under the rack-aware model. We devise two repair schemes to reduce the repair bandwidth for Tamo-Barg codes under the rack-aware model, by setting each repair set as a rack. The first repair scheme provides optimal repair bandwidth for one rack erasure. We then establish a cut-set bound for locally repairable codes under the rack-aware model. Using this bound we show that our second repair scheme is optimal. Furthermore, we consider the partial-repair problem for locally repairable codes under the rack-aware model, and introduce both repair schemes and bounds for this scenario.

In this paper, we discuss codes for distributed storage systems with hierarchical repair properties. Specifically, we devote attention to the repair problem of the rack-aware storage model with locality, aiming to enhance the system's ability to repair a small number of erasures within each rack by locality and efficiently handling a rack erasure with a small repair bandwidth. By employing the regenerating coding technique, we construct a family of array codes with$(r,\ u-r+1)$-locality, where the$u$nodes of each repair set are systematically organized into a rack. When the number of failures is less than$u-r+1$, these failures can be repaired without counting the system bandwidth. In cases where the number of failures exceeds the locality, the failed nodes within a single rack can be recovered with optimal cross-rack bandwidth.

The partially cooperative repair model is an available technology to deal with multiple node failures in a distributed storage system, which does not depend on the heavy assumption of exchanging data with all new nodes, increasing the flexibility of the system. In this paper, constructions and repair schemes for codes with minimum storage overhead and optimal repair bandwidth, i.e., minimum storage partially cooperative regenerating codes are proposed. These constructions show that the theoretic bound on the repair bandwidth of the partially cooperative repair model is tight for the minimum storage cases. Notably, these codes with optimal repair bandwidth may have smaller sub-packetization compared with the known codes for this model.

In this paper, we address the node repair problem of Reed-Solomon (RS) coded distributed storage systems. Specifically, to overcome the challenges of multiple-node failures of RS codes under the rack-aware storage model, we employ good polynomials to guide the placement of the conventional RS codes into racks and then propose a novel repair framework for the resultant rack-aware RS codes, which can transform its repair to that under the homogeneous storage model. As applications of our repair framework, firstly we present the repair scheme of multiple-node failures for some existing constructions, which only have non-trivial solutions for repairing a single-node failure before. Secondly, we deduce several new constructions of rack-aware RS codes supporting the repair of multiple-node failures within a single rack and across multiple racks respectively.

High-resolution dense prediction enables many appealing real-world applications, such as computational photography, autonomous driving, etc. However, the vast computational cost makes deploying state-of-the-art high-resolution dense prediction models on hardware devices difficult. This work presents EfficientViT, a new family of high-resolution vision models with novel lightweight multi-scale attention. Unlike prior high-resolution dense prediction models that rely on heavy self-attention, hardware-inefficient large-kernel convolution, or complicated topology structure to obtain good performances, our lightweight multi-scale attention achieves a global receptive field and multi-scale learning (two critical features for high-resolution dense prediction) with only lightweight and hardware-efficient operations. As such, EfficientViT delivers remarkable performance gains over previous state-of-the-art high-resolution dense prediction models with significant speedup on diverse hardware platforms, including mobile CPU, edge GPU, and cloud GPU. Without performance loss on Cityscapes, our EfficientViT provides up to 8.8× and 3.8× GPU latency reduction over SegFormer and SegNeXt, respectively. For super-resolution, EfficientViT provides up to 6.4× speedup over Restormer while providing 0.11dB gain in PSNR.

We prove a new lower bound on the field size of locally repairable codes (LRCs). Additionally, we construct maximally recoverable (MR) codes which are cyclic. While a known construction for MR codes has the same parameters, it produces non-cyclic codes. Furthermore, we prove both necessary conditions and sufficient conditions that specify when the known non-cyclic MR codes may be permuted to become cyclic, thus proving our construction produces cyclic MR codes with new parameters. Furthermore, using our new bound on the field size, we show that the new cyclic MR codes have optimal field size in certain cases. Other known LRCs are also shown to have optimal field size in certain cases.

The cooperative repair model is an available technology to deal with multiple node failures in distributed storage systems. Recently, explicit constructions of cooperative MSR codes were given by Ye (IEEE Transactions on Information Theory, 2020) with sub-packetization level$(d-k+h)(d-k+1)^{n}$. Specifically, the sub-packetization level is$(h+1)2^{n}$when$d=k+1$. In this paper, we propose a new cooperative repair scheme by means of the inter-instance pairing and intra-instance pairing inherited from the perfect code which reduces the sub-packetization to$2^{n}$when$(h+1)|2^{n}$and$(2\ell +1)2^{n}$when$h+1=(2\ell +1)2^{m}$for$\ell \ge 1$,$m\ge 0$with$d=k+1$helper nodes. That is to say, the sub-packetization is$h + 1 $times or$2^{m}$times less than Ye’s. It turned out to be the best result so far known.

Pose estimation plays a critical role in human-centered vision applications. However, it is difficult to deploy state-of-the-art HRNet-based pose estimation models on resource-constrained edge devices due to the high computational cost (more than 150 GMACs per frame). In this paper, we study efficient architecture design for real-time multi-person pose estimation on edge. We reveal that HRNet's high-resolution branches are redundant for models at the low-computation region via our gradual shrinking experiments. Removing them improves both efficiency and performance. Inspired by this finding, we design LitePose, an efficient single-branch architecture for pose estimation, and introduce two simple approaches to enhance the capacity of LitePose, including fusion deconv head and large kernel conv. On mobile platforms, LitePose reduces the latency by up to$5.0\times$without sacrificing performance, compared with prior state-of-the-art efficient pose estimation models, pushing the frontier of real-time multi-person pose estimation on edge. Our code and pretrained models are released at https://github.com/mit-han-lab/litepose.

We prove a new lower bound on the field size of locally repairable codes (LRCs). Additionally, we construct maximally recoverable (MR) codes which are cyclic. While a known construction for MR codes has the same parameters, it produces non-cyclic codes. Furthermore, we prove necessary and sufficient conditions that specify when the known non-cyclic MR codes may be permuted to become cyclic, thus proving our construction produces cyclic MR codes with new parameters. Furthermore, using our new bound on the field size, we show that the new cyclic MR codes have optimal field size in certain cases. Other known LRCs are also shown to have optimal field size in certain cases.

We study the Singleton-type bound that provides an upper limit on the minimum distance of locally repairable codes. We present an improved bound by carefully analyzing the combinatorial structure of the repair sets. Thus, we show the previous bound is unachievable for certain parameters. We then also provide explicit constructions of optimal codes which show that for certain parameters the new bound is sharp. Additionally, as a byproduct, some previously known codes are shown to attain the new bound and are thus proved to be optimal.

We construct maximally recoverable codes (corresponding to partial MDS codes) which are based on linearized Reed-Solomon codes. The new codes have a smaller field size requirement compared with known constructions. For certain asymptotic regimes, the constructed codes have order-optimal alphabet size, asymptotically matching the known lower bound.

Deep neural networks (DNNs) have achieved unprecedented success in the field of artificial intelligence (AI), including computer vision, natural language processing, and speech recognition. However, their superior performance comes at the considerable cost of computational complexity, which greatly hinders their applications in many resource-constrained devices, such as mobile phones and Internet of Things (IoT) devices. Therefore, methods and techniques that are able to lift the efficiency bottleneck while preserving the high accuracy of DNNs are in great demand to enable numerous edge AI applications. This article provides an overview of efficient deep learning methods, systems, and applications. We start from introducing popular model compression methods, including pruning, factorization, quantization, as well as compact model design. To reduce the large design cost of these manual solutions, we discuss the AutoML framework for each of them, such as neural architecture search (NAS) and automated pruning and quantization. We then cover efficient on-device training to enable user customization based on the local data on mobile devices. Apart from general acceleration techniques, we also showcase several task-specific accelerations for point cloud, video, and natural language processing by exploiting their spatial sparsity and temporal/token redundancy. Finally, to support all these algorithmic advancements, we introduce the efficient deep learning system design from both software and hardware perspectives.

The Singleton-type bound that provides an upper limit on the minimum distance of locally repairable codes is studied. An improved bound is presented by carefully analyzing the combinatorial structure of the repair sets. Thus, we show the previous bound is unachievable for certain parameters. Additionally, as a byproduct, some previously known codes are shown to attain the new bound and are thus proved to be optimal.

Tiny deep learning on microcontroller units (MCUs) is challenging due to the limited memory size. We find that the memory bottleneck is due to the imbalanced memory distribution in convolutional neural network (CNN) designs: the first several blocks have an order of magnitude larger memory usage than the rest of the network. To alleviate this issue, we propose a generic patch-by-patch inference scheduling, which operates only on a small spatial region of the feature map and significantly cuts down the peak memory. However, naive implementation brings overlapping patches and computation overhead. We further propose receptive field redistribution to shift the receptive field and FLOPs to the later stage and reduce the computation overhead. Manually redistributing the receptive field is difficult. We automate the process with neural architecture search to jointly optimize the neural architecture and inference scheduling, leading to MCUNetV2. Patch-based inference effectively reduces the peak memory usage of existing networks by4-8×. Co-designed with neural networks, MCUNetV2 sets a record ImageNetaccuracy on MCU (71.8%) and achieves >90% accuracy on the visual wake words dataset under only 32kB SRAM. MCUNetV2 also unblocks object detection on tiny devices, achieving 16.9% higher mAP on Pascal VOC compared to the state-of-the-art result. Our study largely addressed the memory bottleneck in tinyML and paved the way for various vision applications beyond image classification.

Optimal locally repairable codes with information locality are considered. Optimal codes are constructed, whose length is also order-optimal with respect to a new bound on the code length derived in this article. The length of the constructed codes is super-linear in the alphabet size, which improves upon the well known pyramid codes, whose length is only linear in the alphabet size. The recoverable erasure patterns are also analyzed for the new codes. Based on the recoverable erasure patterns, we construct generalized sector-disk (GSD) codes, which can recover from disk erasures mixed with sector erasures in a more general setting than known sector-disk (SD) codes. Additionally, the number of sectors in the constructed GSD codes is super-linear in the alphabet size, compared with known SD codes, whose number of sectors is only linear in the alphabet size.

Transformers are ubiquitous in Natural Language Processing (NLP) tasks, but they are difficult to be deployed on hardware due to the intensive computation.To enable low-latency inference on resource-constrained hardware platforms, we propose to design Hardware-Aware Transformers (HAT) with neural architecture search.We first construct a large design space with arbitrary encoder-decoder attention and heterogeneous layers.Then we train a Super-Transformer that covers all candidates in the design space, and efficiently produces many SubTransformers with weight sharing.Finally, we perform an evolutionary search with a hardware latency constraint to find a specialized SubTransformer dedicated to run fast on the target hardware.Extensive experiments on four machine translation tasks demonstrate that HAT can discover efficient models for different hardware (CPU, GPU, IoT device).When running WMT'14 translation task on Raspberry Pi-4, HAT can achieve 3× speedup, 3.7× smaller size over baseline Transformer; 2.7× speedup, 3.6× smaller size over Evolved Transformer with 12,041× less search cost and no performance loss.HAT is open-sourced.Elastic Layer Num in Encoder Elastic Head Num (Self Attention) Elastic Hidden Dim in FFN Encoder Layer 2 Encoder Layer m Elastic Embedding Dim Elastic Head Num (Self Attention) Elastic Hidden Dim in FFN Elastic Embedding Dim Elastic Head Num (En-Decoder Attention) Decoder Layer n Elastic Layer Num in Decoder Arbitrary Encoder-Decoder Attention concat ❶ Train a SuperTransformer by uniformly sampling SubTransformers with weight sharing 12/3/

We present APQ, a novel design methodology for efficient deep learning deployment. Unlike previous methods that separately optimize the neural network architecture, pruning policy, and quantization policy, we design to optimize them in a joint manner. To deal with the larger design space it brings, we devise to train a quantization-aware accuracy predictor that is fed to the evolutionary search to select the best fit. Since directly training such a predictor requires time-consuming quantization data collection, we propose to use predictor-transfer technique to get the quantization-aware predictor: we first generate a large dataset of 〈NN architecture, ImageNet accuracy〉 pairs by sampling a pretrained unified once-for-all network and doing direct evaluation; then we use these data to train an accuracy predictor without quantization, followed by transferring its weights to train the quantization-aware predictor, which largely reduces the quantization data collection time. Extensive experiments on ImageNet show the benefits of this joint design methodology: the model searched by our method maintains the same level accuracy as ResNet34 8-bit model while saving 8× BitOps; we achieve 2×/1.3× latency/energy saving compared to MobileNetV2+HAQ [30, 36] while obtaining the same level accuracy; the marginal search cost ofjoint optimization for a new deployment scenario outperforms separate optimizations using ProxylessNAS+AMC+HAQ [5, 12, 36] by 2.3% accuracy while reducing orders of magnitude GPU hours and CO2emission with respect to the training cost.

Optimal locally repairable codes with information locality are considered. Optimal codes are constructed, whose length is also order-optimal with respect to a new bound on the code length derived in this paper. The length of the constructed codes is super-linear in the alphabet size, which improves upon the well known pyramid codes, whose length is only linear in the alphabet size. The recoverable erasure patterns are also analyzed for the new codes. Based on the recoverable erasure patterns, we construct generalized sector-disk (GSD) codes, which can recover from disk erasures mixed with sector erasures in a more general setting than known sector-disk (SD) codes. Additionally, the number of sectors in the constructed GSD codes is superlinear in the alphabet size, compared with known SD codes, whose number of sectors is only linear in the alphabet size.

Efficient on-device learning requires a small memory footprint at training time to fit the tight memory constraint. Existing work solves this problem by reducing the number of trainable parameters. However, this doesn't directly translate to memory saving since the major bottleneck is the activations, not parameters. In this work, we present Tiny-Transfer-Learning (TinyTL) for memory-efficient on-device learning. TinyTL freezes the weights while only learns the memory-efficient bias modules, thus no need to store the intermediate activations. To maintain the adaptation capacity, we introduce a new memory-efficient bias module, the lite residual module, to refine the feature extractor by learning small residual feature maps adding only 3.8% memory overhead. Extensive experiments show that TinyTL significantly saves the memory (up to 6.5x) with little accuracy loss compared to fine-tuning the full network. Compared to fine-tuning the last layer, TinyTL provides significant accuracy improvements (up to 33.8%) with little memory overhead. Furthermore, combined with feature extractor adaptation, TinyTL provides 7.5-12.9x memory saving without sacrificing accuracy compared to fine-tuning the full Inception-V3. Code is released at https://github.com/mit-han-lab/tinyML/tree/master/tinyTL.

Minimal multicast networks are fascinating and efficient combinatorial objects, where the removal of a single link makes it impossible for all receivers to obtain all messages. We study the structure of such networks, and prove some constraints on their possible solutions. We then focus on the combination network, which is one of the simplest and most insightful network in network-coding theory. Of particular interest are minimal combination networks. We study the gap in alphabet size between vector-linear and scalar-linear network-coding solutions for such minimal combination networks and some of their sub-networks. For minimal multicast networks with two source messages we find the maximum possible gap. We define and study sub-networks of the combination network, which we call Kneser networks, and prove that they attain the upper bound on the gap with equality. We also prove that the study of this gap may be limited to the study of sub-networks of minimal combination networks, by using graph homomorphisms connected with the q -analog of Kneser graphs. Additionally, we prove a gap for minimal multicast networks with three or more source messages by studying Kneser networks. Finally, an upper bound on the gap for full minimal combination networks shows nearly no gap, or none in some cases. This is obtained using an MDS-like bound for subspaces over a finite field.

In this paper, locally repairable codes which have optimal minimum Hamming distance with respect to the bound presented by Prakash et al. are considered. New upper bounds on the length of such optimal codes are derived. The new bounds apply to more general cases, and have weaker requirements compared with the known ones. In this sense, they both improve and generalize previously known bounds. Further, optimal codes are constructed, whose length is order-optimal with respect to the new upper bounds. Notably, the length of the codes is super-linear in the alphabet size.

Locally repairable codes are desirable for distributed storage systems to improve the repair efficiency. In this paper, a new combination of codes with locality and codes with multiple disjoint repair sets (also called availability) is introduced. Accordingly, a Singleton-type bound is derived for the new code, which contains those bounds in [9], [20], [28] as special cases. Optimal constructions are proposed with respect to the new bound. In addition, these constructions can also generate optimal codes with multiple disjoint repair sets with respect to the bound in [28], which to the best of our knowledge, are the first explicit constructions that can achieve the bound in [28].

Reinforcement learning (RL) has recently been introduced to interactive recommender systems (IRS) because of its nature of learning from dynamic interactions and planning for long-run performance. As IRS is always with thousands of items to recommend (i.e., thousands of actions), most existing RL-based methods, however, fail to handle such a large discrete action space problem and thus become inefficient. The existing work that tries to deal with the large discrete action space problem by utilizing the deep deterministic policy gradient framework suffers from the inconsistency between the continuous action representation (the output of the actor network) and the real discrete action. To avoid such inconsistency and achieve high efficiency and recommendation effectiveness, in this paper, we propose a Tree-structured Policy Gradient Recommendation (TPGR) framework, where a balanced hierarchical clustering tree is built over the items and picking an item is formulated as seeking a path from the root to a certain leaf of the tree. Extensive experiments on carefully-designed environments based on two real-world datasets demonstrate that our model provides superior recommendation performance and significant efficiency improvement over state-of-the-art methods.

The combination network is one of the simplest and insightful networks in coding theory. The vector network coding solutions for this network and some of its sub-networks are examined. For a fixed alphabet size of a vector network coding solution, an upper bound on the number of nodes in the network is obtained. This bound is an MDS bound for subspaces over a finite field. A family of sub-networks of combination networks is defined. It is proved that for this family of networks, which are minimal multicast networks, there is a gap in the minimum alphabet size between vector network coding solutions and scalar network coding solutions. This gap is obtained for any number of messages and is based on coloring of the q-Kneser graph and a new hypergraph generalization for it.

Optimal locally repairable codes with respect to the bound presented by Prakash et al. are considered. New upper bounds on the length of such optimal codes are derived. The new bounds both improve and generalize previously known bounds. Optimal codes are constructed, whose length is order optimal when compared with the new upper bounds. The length of the codes is super linear in the alphabet size.

Locally repairable codes are desirable for distributed storage systems to improve the repair efficiency. In this paper, a connection between locally repairable codes with multiple disjoint repair sets and packings is derived under the condition that each repair set contains exactly one check symbol. Particularly, conditions under which an optimal locally repairable code corresponds to a packing are also characterized. As an application of this connection, some optimal locally repairable codes can be obtained by packings. Specifically, two constructions of locally repairable codes are proposed which not only generalize some known explicit constructions but also give optimal locally repairable codes with flexible new parameters.

Techniques for automatically designing deep neural network architectures such as reinforcement learning based approaches have recently shown promising results. However, their success is based on vast computational resources (e.g. hundreds of GPUs), making them difficult to be widely used. A noticeable limitation is that they still design and train each network from scratch during the exploration of the architecture space, which is highly inefficient. In this paper, we propose a new framework toward efficient architecture search by exploring the architecture space based on the current network and reusing its weights. We employ a reinforcement learning agent as the meta-controller, whose action is to grow the network depth or layer width with function-preserving transformations. As such, the previously validated networks can be reused for further exploration, thus saves a large amount of computational cost. We apply our method to explore the architecture space of the plain convolutional neural networks (no skip-connections, branching etc.) on image benchmark datasets (CIFAR-10, SVHN) with restricted computational resources (5 GPUs). Our method can design highly competitive networks that outperform existing networks using the same design scheme. On CIFAR-10, our model without skip-connections achieves 4.23% test error rate, exceeding a vast majority of modern architectures and approaching DenseNet. Furthermore, by applying our method to explore the DenseNet architecture space, we are able to achieve more accurate networks with fewer parameters.

Automatically generating coherent and semantically meaningful text has many applications in machine translation, dialogue systems, image captioning, etc. Recently, by combining with policy gradient, Generative Adversarial Nets(GAN) that use a discriminative model to guide the training of the generative model as a reinforcement learning policy has shown promising results in text generation. However, the scalar guiding signal is only available after the entire text has been generated and lacks intermediate information about text structure during the generative process. As such, it limits its success when the length of the generated text samples is long (more than 20 words). In this paper, we propose a new framework, called LeakGAN, to address the problem for long text generation. We allow the discriminative net to leak its own high-level extracted features to the generative net to further help the guidance. The generator incorporates such informative signals into all generation steps through an additional MANAGER module, which takes the extracted features of current generated words and outputs a latent vector to guide the WORKER module for next-word generation.Our extensive experiments on synthetic data and various real-world tasks with Turing test demonstrate that LeakGAN is highly effective in long text generation and also improves the performance in short text generation scenarios. More importantly, without any supervision, LeakGAN would be able to implicitly learn sentence structures only through the interaction between MANAGER and WORKER.

We introduce MAgent, a platform to support research and development of many-agent reinforcement learning. Unlike previous research platforms on single or multi-agent reinforcement learning, MAgent focuses on supporting the tasks and the applications that require hundreds to millions of agents. Within the interactions among a population of agents, it enables not only the study of learning algorithms for agents' optimal polices, but more importantly, the observation and understanding of individual agent's behaviors and social phenomena emerging from the AI society, including communication languages, leaderships, altruism. MAgent is highly scalable and can host up to one million agents on a single GPU server. MAgent also provides flexible configurations for AI researchers to design their customized environments and agents. In this demo, we present three environments designed on MAgent and show emerged collective intelligence by learning from scratch.

We introduce a new function-preserving transformation for efficient neural architecture search. This network transformation allows reusing previously trained networks and existing successful architectures that improves sample efficiency. We aim to address the limitation of current network transformation operations that can only perform layer-level architecture modifications, such as adding (pruning) filters or inserting (removing) a layer, which fails to change the topology of connection paths. Our proposed path-level transformation operations enable the meta-controller to modify the path topology of the given network while keeping the merits of reusing weights, and thus allow efficiently designing effective structures with complex path topologies like Inception models. We further propose a bidirectional tree-structured reinforcement learning meta-controller to explore a simple yet highly expressive tree-structured architecture space that can be viewed as a generalization of multi-branch architectures. We experimented on the image classification datasets with limited computational resources (about 200 GPU-hours), where we observed improved parameter efficiency and better test results (97.70% test accuracy on CIFAR-10 with 14.3M parameters and 74.6% top-1 accuracy on ImageNet in the mobile setting), demonstrating the effectiveness and transferability of our designed architectures.

Global Navigation Satellite System (GNSS) radio occultation (RO) has been widely used in the prediction of weather, climate, and space weather, particularly in the area of tropospheric analyses. However, one of the issues with GNSS RO measurements is that they are interfered with by the signals reflected from the earth's surface. Many RO events are subject to such interfered GNSS measurements, which are considerably difficult to extract from the GNSS RO measurements. To precisely identify interfered RO events, an improved machine learning approach-a gradient descent artificial neural network (ANN)-aided radio-holography method-is proposed in this paper. Since this method is more complex than most other machine learning methods, for improving its efficiency through the reduction in computational time for near-real-time applications, a scale factor and a regularization factor are also adjusted in the ANN approach. This approach was validated using Constellation Observing System for Meteorology, Ionosphere, and Climate/FC-3 atmPhs (level 1b) data during the period of day of year 172-202, 2015, and its detection results were compared with the flag data set provided by Radio Occultation Meteorology Satellite Application Facilities for the performance assessment and validation of the new approach. The results were also compared with those of the support vector machine method for improvement assessment. The comparison results showed that the proposed method can considerably improve both the success rate of GNSS RO reflection detection and the computational efficiency.

Programmatic display advertising, which enables advertisers to make real-time decisions on individual ad display opportunities so as to achieve a precise audience marketing, has become a key technique for online advertising. However, the constrained budget setting still restricts unlimited ad impressions. As a result, a smart strategy for ad impression selection is necessary for the advertisers to maximize positive user responses such as clicks or conversions, under the constraints of both ad volume and campaign budget. In this paper, we borrow in the idea of top-N ranking and filtering techniques from information retrieval and propose an effective ad impression volume ranking method for each ad campaign, followed by a sequential selection strategy considering the remaining ad volume and budget, to smoothly deliver the volume filtering while maximizing campaign efficiency. The extensive experiments on two benchmarking datasets and a commercial ad platform demonstrate large performance superiority of our proposed solution over traditional methods, especially under tight budgets.

The majority of online display ads are served through real-time bidding (RTB) --- each ad display impression is auctioned off in real-time when it is just being generated from a user visit. To place an ad automatically and optimally, it is critical for advertisers to devise a learning algorithm to cleverly bid an ad impression in real-time. Most previous works consider the bid decision as a static optimization problem of either treating the value of each impression independently or setting a bid price to each segment of ad volume. However, the bidding for a given ad campaign would repeatedly happen during its life span before the budget runs out. As such, each bid is strategically correlated by the constrained budget and the overall effectiveness of the campaign (e.g., the rewards from generated clicks), which is only observed after the campaign has completed. Thus, it is of great interest to devise an optimal bidding strategy sequentially so that the campaign budget can be dynamically allocated across all the available impressions on the basis of both the immediate and future rewards. In this paper, we formulate the bid decision process as a reinforcement learning problem, where the state space is represented by the auction information and the campaign's real-time parameters, while an action is the bid price to set. By modeling the state transition via auction competition, we build a Markov Decision Process framework for learning the optimal bidding policy to optimize the advertising performance in the dynamic real-time bidding environment. Furthermore, the scalability problem from the large real-world auction volume and campaign budget is well handled by state value approximation using neural networks. The empirical study on two large-scale real-world datasets and the live A/B testing on a commercial platform have demonstrated the superior performance and high efficiency compared to state-of-the-art methods.

As an optimal combinatorial object, zero-difference balanced (ZDB) functions introduced by Ding in 2008, are a generalization of the well-known perfect nonlinear functions. ZDB functions have received much attention in recent years due to its important applications in coding theory and sequence design. One objective of this paper is to present a construction of ZDB functions based on a kind of generalized cyclotomy. It generates ZDB functions over cyclic group with new parameters which can not be produced by earlier constructions. Another objective of this paper is to employ these ZDB functions to obtain at the same time: 1) optimal constant-composition codes; 2) perfect difference systems of sets; and 3) optimal frequency-hopping sequences, all with new parameters.

Predicting user responses, such as clicks and conversions, is of great importance and has found its usage inmany Web applications including recommender systems, websearch and online advertising. The data in those applicationsis mostly categorical and contains multiple fields, a typicalrepresentation is to transform it into a high-dimensional sparsebinary feature representation via one-hot encoding. Facing withthe extreme sparsity, traditional models may limit their capacityof mining shallow patterns from the data, i.e. low-order featurecombinations. Deep models like deep neural networks, on theother hand, cannot be directly applied for the high-dimensionalinput because of the huge feature space. In this paper, we proposea Product-based Neural Networks (PNN) with an embeddinglayer to learn a distributed representation of the categorical data, a product layer to capture interactive patterns between interfieldcategories, and further fully connected layers to explorehigh-order feature interactions. Our experimental results on twolarge-scale real-world ad click datasets demonstrate that PNNsconsistently outperform the state-of-the-art models on various metrics.

Frequency-hopping sequences (FHSs) with favorable partial Hamming correlation properties are desirable in many synchronization and multiple-access systems. An FHS set is said to be strictly optimal if it has optimal partial Hamming correlation for any correlation window. In this paper, we derive upper bounds on the family sizes of FHS sets with respect to partial Hamming correlation from some classical bounds on error-correcting codes. We then present strictly optimal FHS sets having optimal family sizes with respect to one of the new bounds. In particular, our construction gives new parameters not covered in the literature.

Frequency-hopping sequences (FHSs) with favorable partial Hamming correlation properties have important applications in many synchronization and multiple-access systems. Strictly optimal FHSs are those FHSs with optimal partial Hamming autocorrelation irrespective of the correlation window length. In this paper, strictly optimal FHSs are investigated from a combinatorial approach. A generic connection between strictly optimal FHSs and disjoint cyclic perfect Mendelsohn difference families is established. By virtue of this connection, new strictly optimal FHSs are generated from some disjoint CPMDFs. These strictly optimal FHSs have new parameters not covered in the literature.

In this paper, we focus on a mobile wireless network comprising a powerful communication center and a multitude of mobile users. We investigate the propagation of latency-constrained content in the wireless network characterized by heterogeneous (time-varying and user-dependent) wireless channel conditions, heterogeneous user mobility, and where communication could occur in a hybrid format (e.g., directly from the central controller or by exchange with other mobiles in a peer-to-peer manner). We show that exploiting double opportunities, i.e., both time-varying channel conditions and mobility, can result in substantial performance gains. We develop a class of double opportunistic multicast schedulers and prove their optimality in terms of both utility and fairness under heterogeneous channel conditions and user mobility. Extensive simulation results are provided to demonstrate that these algorithms can not only substantially boost the throughput of all users (e.g., by 50% to 150%), but also achieve different consideration of fairness among individual users and groups of users.

Optical orthogonal codes (OOCs) are widely used as spreading codes in optical fiber networks. In this paper, a bound on the code size of 2-D OOCs with both at most one-pulse per wavelength (AM-OPPW) and at most one-pulse per time slot (AM-OPPTS) is derived. Accordingly, two constructions of optimal 2-D OOCs with both AM-OPPW and AM-OPPTS are proposed via the generalized cyclotomic classes. Furthermore, optimal 2-D OOC with AM-OPPW can be also constructed by adding more codewords into the 2-D OOCs with both AM-OPPW and AM-OPPTS.

Frequency-hopping sequences (FHSs) with favorable partial Hamming correlation properties have important applications in many synchronization and multiple-access systems. In this paper, lower bounds on the partial Hamming correlation of FHSs and FHS sets are proposed. They slightly improve the known bounds by Eun et al. and Zhou et al. A construction of FHSs and FHS sets having optimal partial Hamming correlation with respect to the improved bounds is also presented based on the theory of generalized cyclotomy. Our construction yields optimal FHSs and FHS sets with new and flexible parameters not covered in this paper.

In this paper, we focus on mobile wireless networks comprising of a powerful communication center and a multitude of mobile users. We investigate the propagation of deadline-based content in the wireless network characterized by heterogeneous (time-varying and user-dependent) wireless channel conditions, heterogeneous user mobility, and where communication could occur in a hybrid format (e.g., directly from the central controller or by exchange with other mobiles in a peer-to-peer manner). We show that exploiting double opportunities, i.e., both time-varying channel conditions and mobility, can result in substantial performance gains. We develop a class of double opportunistic multicast schedulers and prove their optimality in terms of both utility and fairness under heterogeneous channel conditions and user mobility. Extensive simulation results are provided to demonstrate that these algorithms can not only substantially boost the throughput of all users (e.g., by 50% to 150%), but also achieve different consideration of fairness among individual users and groups of users.

In this paper, a new construction of zero-difference balanced functions defined on is given, where is an odd positive integer. Based on the generic constructions proposed by Ding, optimal constant composition codes and perfect difference systems of sets with new parameters can be generated from the zero-difference balanced functions constructed in this paper.

In this paper, a new generalized cyclotomy with respect to a positive odd integer is introduced, and a construction of frequency hopping sequence sets and two constructions of frequency hopping sequences are proposed as its applications. The frequency hopping sequence sets and frequency hopping sequences obtained in this paper can be optimal with respect to the Peng-Fan bound and Lempel-Greenberger bound, respectively. Further, the length of sequences in the optimal frequency hopping sequence sets can be any odd integer larger than 3. Some of them have new parameters.

In this paper, we propose an interleaving construction of new sets of frequency hopping sequences from the known ones. By choosing suitable known optimal frequency hopping sequences and sets of frequency hopping sequences and then recursively applying the proposed construction, optimal frequency hopping sequences and sets of frequency hopping sequences with new parameters can be obtained.

The study in mobile ad-hoc networks (MANET) is facing challenges brought by recent discovery of non-exponential behavior of the inter-contact time distribution of mobile nodes. In this paper, we analyze various characteristics of the relative mobility of a random pair of nodes in MANET to show that they produce inter-contact time with different aging properties. First, by fixing one node and resorting to the random walks on directed graphs, we mathematically prove that under four classes of stochastic mobility patterns, the resulting inter-contact times have constant/decreasing/increasing failure rate and new-better-than-used property. Then, we consider the case when both nodes are mobile and use simulation results to uncover the aging property of their inter-contact times under random waypoint models and random walk mobility models. This aging property tells us how to correctly relate the past experience of mobile nodes with their future behavior, thereby allowing tremendous opportunities brought by the memory structure in the non-exponential inter contact time, which would be impossible under the widely assumed exponentially distributed (memoryless) inter-contact time. As an application of our results, we establish for the first time that the approach based on exponential inter-contact time assumption can either under-estimate or over-estimate the actual system performance, under different stochastic mobility patterns indexed by their aging properties. Our results on aging properties also provide theoretic guidelines on how to exploit the memory structure toward better design of protocols under general mobility.

Multicast is an efficient means of transmitting the same content to multiple receivers while minimizing network resource usage. Applications that can benefit from multicast such as multimedia streaming and download, are now being deployed over 3G wireless data networks. Existing multicast schemes transmit data at a fixed rate that can accommodate the farthest located users in a cell. However, users belonging to the same multicast group can have widely different channel conditions. Thus existing schemes are too conservative by limiting the throughput of users close to the base station. We propose two proportional fair multicast scheduling algorithms that can adapt to dynamic channel states in cellular data networks that use time division multiplexing: inter-group proportional fairness (IPF) and multicast proportional fairness (MPF). These scheduling algorithms take into account (1) reported data rate requests from users which dynamically change to match their link states to the base station, and (2) the average received throughput of each user inside its cell. This information is used by the base station to select an appropriate data rate for each group. We prove that IPF and MPF achieve proportional fairness among groups and among all users inside a cell respectively. Through extensive packet-level simulations, we demonstrate that these algorithms achieve good balance between throughput and fairness among users and groups.

Scheduling algorithms assign contention probability for each link in wireless ad hoc networks and plays a key role in deciding the system performance. Recently, many low-cost distributed scheduling algorithms are proposed. In this paper, we propose to improve the performance of a class of distributed collision-based scheduling algorithms, called constant-time distributed scheduling algorithms, by exploring the advantage brought by the unevenness of links' contention probabilities. Specifically, we prove that there exists ordering relationship for the success probability of any neighborhood when the contention probability vectors are ordered in the sense of majorization. We show how to modify the existing algorithms so as to find a new contention probability vector that majorizes the original one in a distributed manner. Our simulation results indicate that by using our modified algorithms, the average queue-lengths of a stable system can be reduced by 25% to 50%, while the capacity region remains the same. Our modification to the existing algorithms is extremely simple and entails essentially zero additional cost.

The class of isotropic random walk mobility models, including random direction mobility model, random walk mobility model and Brownian motion mobility model, has been widely used in the study of Mobile Ad-Hoc Networks for mobility modeling and control. In this paper, we show an important property for contact time of isotropic random walk mobility models. Specifically, we find that the mean contact time of two mobile nodes following isotropic random walk mobility models is invariant with respect to the step-length distribution under both the simplest distance-based (Boolean) interference model and the more realistic SINR-based interference model. We also study the effect of system parameters on the contact and inter-meeting time of mobile nodes and discuss their higher-order statistics.

Recent discovery of the mixture (power-law and exponential) behavior of inter-meeting time distribution of mobile nodes presents new challenge to the problem of mobility modeling and its effect on the network performance. Existing studies on this problem via the average inter-meeting time become insufficient when the inter-meeting time distribution starts to deviate from exponential one. This insufficiency necessarily leads to the increasing difficulty in the performance analysis of forwarding algorithms in mobile ad-hoc networks (MANET). In this paper, we analyze the effect of mobility patterns on the inter-meeting time distribution. We first identify the critical timescale in the inter-meeting distribution, at which the transition from power-law to exponential takes place, in terms of the domain size and the statistics of the mobility pattern. We then prove that stronger correlations in mobility patterns lead to heavier (non-exponential) 'head' of the inter-meeting time distribution. We also prove that there exists an invariance property for several contact-based metrics such as inter-meeting, contact, inter-any-contact time under both distance-based (Boolean) and physical interference (SINR) based models, in that the averages of those contact-based metrics do not depend on the degree of correlations in the mobility patterns. Our results collectively suggest a convex ordering relationship among inter-meeting times of various mobility models indexed by their degrees of correlation, which is in good agreement with the ordering of network performance under a set of mobility patterns whose inter-meeting time distributions have power-law 'head' followed by exponential 'tail'.

Window growth function for congestion control is a strong determinant of protocol behaviors, especially its second and higher-order behaviors associated with the distribution of transmission rates, its variances, and protocol stability. This paper presents a new stochastic tool, called convex ordering, that provides an ordering of any convex function of transmission rates of two protocols and valuable insights into high order behaviors of protocols. As the ordering determined by this tool is consistent with any convex function of rates, it can be applied to any unknown metric for protocol performance that consists of some high-order moments of transmission rates, as well as those already known such as rate variance. Using the tool, it is analyzed that a protocol with a growth function that starts off with a concave function and then switches to a convex function (e.g., an odd order function such as x3and x5) around the maximum window size in the previous loss epoch, gives the smallest rate variation under a variety of network conditions. Among existing protocols, BIC and CUBIC have this window growth function. Experimental and simulation results confirm the analytical findings.

Multicast is an efficient means of transmitting the same content to multiple receivers while minimizing network resource usage. Applications that can benefit from multicast such as multimedia streaming and download, are now being deployed over 3G wireless data networks. Existing multicast schemes transmit data at a fixed rate that can accommodate the farthest located users in a cell. However, users belonging to the same multicast group can have widely different channel conditions. Thus existing schemes are too conservative by limiting the throughput of users close to the base station. We propose two proportional fair multicast scheduling algorithms that can adapt to dynamic channel states in cellular data networks that use time division multiplexing: Inter-group Proportional Fairness (IPF) and multicast proportional fairness (MPF). These scheduling algorithms take into account (1) reported data rate requests from users which dynamically change to match their link states to the base station, and (2) the average received throughput of each user inside its cell. This information is used by the base station to select an appropriate data rate for each group. We prove that IPF and MPF achieve proportional fairness among groups and among all users in a group inside a cell respectively. Through extensive packet-level simulations, we demonstrate that these algorithms achieve good balance between throughput and fairness among users and groups.

Inter-meeting time between mobile nodes is one of the key metrics in a Mobile Ad-hoc Network (MANET) and central to the end-to-end delay and forwarding algorithms. It is typically assumed to be exponentially distributed in many performance studies of MANET or numerically shown to be exponentially distributed under most existing mobility models in the literature. However, recent empirical results show otherwise: the inter-meeting time distribution in fact follows a power-law. This outright discrepancy potentially undermines our understanding of the performance tradeoffs in MANET obtained under the exponential distribution ofthe inter-meeting time, and thus calls for further study on the power-law inter-meeting time including its fundamental cause, mobility modeling, and its effect. In this paper, we rigorously prove that a finite domain, on which most of the current mobility models are defined, plays an important role in creating the exponential tail of the inter-meeting time. We also prove that by simply removing the boundary in a simple two-dimensional isotropic random walk model, we are able to obtain the empirically observed power-law decay of the inter-meeting time. We then discuss the relationship between the size of the boundary and the relevant time scale of the network scenario under consideration. Our results thus provide guidelines on the design of new mobility models with power-law inter-meeting time distribution, new protocols including packet forwarding algorithms, as well as their performance analysis.