Zebin Yang

Zebin Yang, Yijiahao Qi, Tong Xie et al.

Vision-Language-Action (VLA) models have shown remarkable success in robotic tasks like manipulation by fusing a language model's reasoning with a vision model's 3D understanding. However, their high computational cost remains a major obstacle for real-world applications that require real-time performance. We observe that the actions within a task have varying levels of importance: critical steps demand high precision, while less important ones can tolerate more variance. Leveraging this insight, we propose DySL-VLA, a novel framework that addresses computational cost by dynamically skipping VLA layers based on each action's importance. DySL-VLA categorizes its layers into two types: informative layers, which are consistently executed, and incremental layers, which can be selectively skipped. To intelligently skip layers without sacrificing accuracy, we invent a prior-post skipping guidance mechanism to determine when to initiate layer-skipping. We also propose a skip-aware two-stage knowledge distillation algorithm to efficiently train a standard VLA into a DySL-VLA. Our experiments indicate that DySL-VLA achieves 2.1% improvement in success length over Deer-VLA on the Calvin dataset, while simultaneously reducing trainable parameters by a factor of 85.7 and providing a 3.75x speedup relative to the RoboFlamingo baseline at iso-accuracy. Our code is available on https://github.com/PKU-SEC-Lab/DYSL_VLA.

Renjie Wei, Songqiang Xu, Linfeng Zhong et al.

State space models (SSMs) like Mamba have recently attracted much attention. Compared to Transformer-based large language models (LLMs), Mamba achieves linear computation complexity with the sequence length and demonstrates superior performance. However, Mamba is hard to accelerate due to the scattered activation outliers and the complex computation dependency, rendering existing LLM accelerators inefficient. In this paper, we propose LightMamba that co-designs the quantization algorithm and FPGA accelerator architecture for efficient Mamba inference. We first propose an FPGA-friendly post-training quantization algorithm that features rotation-assisted quantization and power-of-two SSM quantization to reduce the majority of computation to 4-bit. We further design an FPGA accelerator that partially unrolls the Mamba computation to balance the efficiency and hardware costs. Through computation reordering as well as fine-grained tiling and fusion, the hardware utilization and memory efficiency of the accelerator get drastically improved. We implement LightMamba on Xilinx Versal VCK190 FPGA and achieve 4.65x to 6.06x higher energy efficiency over the GPU baseline. When evaluated on Alveo U280 FPGA, LightMamba reaches 93 tokens/s, which is 1.43x that of the GPU baseline. Our code is available at https://github.com/PKU-SEC-Lab/LightMamba.

Zebin Yang, Sunjian Zheng, Tong Xie et al.

Object-goal navigation (ObjNav) tasks an agent with navigating to the location of a specific object in an unseen environment. Embodied agents equipped with large language models (LLMs) and online constructed navigation maps can perform ObjNav in a zero-shot manner. However, existing agents heavily rely on giant LLMs on the cloud, e.g., GPT-4, while directly switching to small LLMs, e.g., LLaMA3.2-11b, suffer from significant success rate drops due to limited model capacity for understanding complex navigation maps, which prevents deploying ObjNav on local devices. At the same time, the long prompt introduced by the navigation map description will cause high planning latency on local devices. In this paper, we propose EfficientNav to enable on-device efficient LLM-based zero-shot ObjNav. To help the smaller LLMs better understand the environment, we propose semantics-aware memory retrieval to prune redundant information in navigation maps. To reduce planning latency, we propose discrete memory caching and attention-based memory clustering to efficiently save and re-use the KV cache. Extensive experimental results demonstrate that EfficientNav achieves 11.1% improvement in success rate on HM3D benchmark over GPT-4-based baselines, and demonstrates 6.7x real-time latency reduction and 4.7x end-to-end latency reduction over GPT-4 planner. Our code will be released soon.

Rongqing Cong, Wenyang He, Mingxuan Li et al.

Large language models (LLMs) with Transformer architectures have become phenomenal in natural language processing, multimodal generative artificial intelligence, and agent-oriented artificial intelligence. The self-attention module is the most dominating sub-structure inside Transformer-based LLMs. Computation using general-purpose graphics processing units (GPUs) inflicts reckless demand for I/O bandwidth for transferring intermediate calculation results between memories and processing units. To tackle this challenge, this work develops a fully customized vanilla self-attention accelerator, AttentionLego, as the basic building block for constructing spatially expandable LLM processors. AttentionLego provides basic implementation with fully-customized digital logic incorporating Processing-In-Memory (PIM) technology. It is based on PIM-based matrix-vector multiplication and look-up table-based Softmax design. The open-source code is available online: https://bonany.cc/attentionleg.

Agus Sudjianto, Aijun Zhang, Zebin Yang et al.

PiML (read $π$-ML, /`pai`em`el/) is an integrated and open-access Python toolbox for interpretable machine learning model development and model diagnostics. It is designed with machine learning workflows in both low-code and high-code modes, including data pipeline, model training and tuning, model interpretation and explanation, and model diagnostics and comparison. The toolbox supports a growing list of interpretable models (e.g. GAM, GAMI-Net, XGB1/XGB2) with inherent local and/or global interpretability. It also supports model-agnostic explainability tools (e.g. PFI, PDP, LIME, SHAP) and a powerful suite of model-agnostic diagnostics (e.g. weakness, reliability, robustness, resilience, fairness). Integration of PiML models and tests to existing MLOps platforms for quality assurance are enabled by flexible high-code APIs. Furthermore, PiML toolbox comes with a comprehensive user guide and hands-on examples, including the applications for model development and validation in banking. The project is available at https://github.com/SelfExplainML/PiML-Toolbox.

Shuzhang Zhong, Zebin Yang, Meng Li et al.

Recent advancements in generative large language models (LLMs) have significantly boosted the performance in natural language processing tasks. However, their efficiency is hampered by the inherent limitations in autoregressive token generation. While parallel decoding with token tree verification, e.g., Medusa, has been proposed to improve decoding parallelism and efficiency, it often struggles with maintaining contextual relationships due to its independent token prediction approach and incurs significant verification overhead, especially with large tree sizes and batch processing. In this paper, we propose ProPD, an efficient LLM parallel decoding framework based on dynamic token tree pruning and generation. ProPD features an advanced early pruning mechanism to efficiently eliminate unpromising token sequences to improve verification efficiency. Additionally, it introduces a dynamic token tree generation algorithm to balance the computation and parallelism of the verification phase in real-time and maximize the overall efficiency across different batch sizes, sequence lengths, and tasks, etc. We verify ProPD across a diverse set of datasets, LLMs, and batch sizes and demonstrate ProPD consistently outperforms existing decoding algorithms by 1.1-3.2x.

Zebin Yang, Renze Chen, Taiqiang Wu et al.

In this paper, we propose MCUBERT to enable language models like BERT on tiny microcontroller units (MCUs) through network and scheduling co-optimization. We observe the embedding table contributes to the major storage bottleneck for tiny BERT models. Hence, at the network level, we propose an MCU-aware two-stage neural architecture search algorithm based on clustered low-rank approximation for embedding compression. To reduce the inference memory requirements, we further propose a novel fine-grained MCU-friendly scheduling strategy. Through careful computation tiling and re-ordering as well as kernel design, we drastically increase the input sequence lengths supported on MCUs without any latency or accuracy penalty. MCUBERT reduces the parameter size of BERT-tiny and BERT-mini by 5.7$\times$ and 3.0$\times$ and the execution memory by 3.5$\times$ and 4.3$\times$, respectively. MCUBERT also achieves 1.5$\times$ latency reduction. For the first time, MCUBERT enables lightweight BERT models on commodity MCUs and processing more than 512 tokens with less than 256KB of memory.

Zebin Yang, Agus Sudjianto, Xiaoming Li et al.

Tree ensemble models like random forests and gradient boosting machines are widely used in machine learning due to their excellent predictive performance. However, a high-performance ensemble consisting of a large number of decision trees lacks sufficient transparency and explainability. In this paper, we demonstrate that when shallow decision trees are used as base learners, the ensemble learning algorithms can not only become inherently interpretable subject to an equivalent representation as the generalized additive models but also sometimes lead to better generalization performance. First, an interpretation algorithm is developed that converts the tree ensemble into the functional ANOVA representation with inherent interpretability. Second, two strategies are proposed to further enhance the model interpretability, i.e., by adding constraints in the model training stage and post-hoc effect pruning. Experiments on simulations and real-world datasets show that our proposed methods offer a better trade-off between model interpretation and predictive performance, compared with its counterpart benchmarks.

Yifeng Guo, Yu Su, Zebin Yang et al.

In recent years, the field of recommendation systems has attracted increasing attention to developing predictive models that provide explanations of why an item is recommended to a user. The explanations can be either obtained by post-hoc diagnostics after fitting a relatively complex model or embedded into an intrinsically interpretable model. In this paper, we propose the explainable recommendation systems based on a generalized additive model with manifest and latent interactions (GAMMLI). This model architecture is intrinsically interpretable, as it additively consists of the user and item main effects, the manifest user-item interactions based on observed features, and the latent interaction effects from residuals. Unlike conventional collaborative filtering methods, the group effect of users and items are considered in GAMMLI. It is beneficial for enhancing the model interpretability, and can also facilitate the cold-start recommendation problem. A new Python package GAMMLI is developed for efficient model training and visualized interpretation of the results. By numerical experiments based on simulation data and real-world cases, the proposed method is shown to have advantages in both predictive performance and explainable recommendation.

Agus Sudjianto, William Knauth, Rahul Singh et al.

The deep neural networks (DNNs) have achieved great success in learning complex patterns with strong predictive power, but they are often thought of as "black box" models without a sufficient level of transparency and interpretability. It is important to demystify the DNNs with rigorous mathematics and practical tools, especially when they are used for mission-critical applications. This paper aims to unwrap the black box of deep ReLU networks through local linear representation, which utilizes the activation pattern and disentangles the complex network into an equivalent set of local linear models (LLMs). We develop a convenient LLM-based toolkit for interpretability, diagnostics, and simplification of a pre-trained deep ReLU network. We propose the local linear profile plot and other visualization methods for interpretation and diagnostics, and an effective merging strategy for network simplification. The proposed methods are demonstrated by simulation examples, benchmark datasets, and a real case study in home lending credit risk assessment.

Zebin Yang, Aijun Zhang

Hyperparameter optimization (HPO) plays a central role in the automated machine learning (AutoML). It is a challenging task as the response surfaces of hyperparameters are generally unknown, hence essentially a global optimization problem. This paper reformulates HPO as a computer experiment and proposes a novel sequential uniform design (SeqUD) strategy with three-fold advantages: a) the hyperparameter space is adaptively explored with evenly spread design points, without the need of expensive meta-modeling and acquisition optimization; b) the batch-by-batch design points are sequentially generated with parallel processing support; c) a new augmented uniform design algorithm is developed for the efficient real-time generation of follow-up design points. Extensive experiments are conducted on both global optimization tasks and HPO applications. The numerical results show that the proposed SeqUD strategy outperforms benchmark HPO methods, and it can be therefore a promising and competitive alternative to existing AutoML tools.

Zebin Yang, Hengtao Zhang, Agus Sudjianto et al.

Network initialization is the first and critical step for training neural networks. In this paper, we propose a novel network initialization scheme based on the celebrated Stein's identity. By viewing multi-layer feedforward neural networks as cascades of multi-index models, the projection weights to the first hidden layer are initialized using eigenvectors of the cross-moment matrix between the input's second-order score function and the response. The input data is then forward propagated to the next layer and such a procedure can be repeated until all the hidden layers are initialized. Finally, the weights for the output layer are initialized by generalized linear modeling. Such a proposed SteinGLM method is shown through extensive numerical results to be much faster and more accurate than other popular methods commonly used for training neural networks.

Zebin Yang, Aijun Zhang, Agus Sudjianto

The lack of interpretability is an inevitable problem when using neural network models in real applications. In this paper, an explainable neural network based on generalized additive models with structured interactions (GAMI-Net) is proposed to pursue a good balance between prediction accuracy and model interpretability. GAMI-Net is a disentangled feedforward network with multiple additive subnetworks; each subnetwork consists of multiple hidden layers and is designed for capturing one main effect or one pairwise interaction. Three interpretability aspects are further considered, including a) sparsity, to select the most significant effects for parsimonious representations; b) heredity, a pairwise interaction could only be included when at least one of its parent main effects exists; and c) marginal clarity, to make main effects and pairwise interactions mutually distinguishable. An adaptive training algorithm is developed, where main effects are first trained and then pairwise interactions are fitted to the residuals. Numerical experiments on both synthetic functions and real-world datasets show that the proposed model enjoys superior interpretability and it maintains competitive prediction accuracy in comparison to the explainable boosting machine and other classic machine learning models.

Zebin Yang, Aijun Zhang, Agus Sudjianto

Prediction accuracy and model explainability are the two most important objectives when developing machine learning algorithms to solve real-world problems. The neural networks are known to possess good prediction performance, but lack of sufficient model interpretability. In this paper, we propose to enhance the explainability of neural networks through the following architecture constraints: a) sparse additive subnetworks; b) projection pursuit with orthogonality constraint; and c) smooth function approximation. It leads to an explainable neural network (xNN) with the superior balance between prediction performance and model interpretability. We derive the necessary and sufficient identifiability conditions for the proposed xNN model. The multiple parameters are simultaneously estimated by a modified mini-batch gradient descent method based on the backpropagation algorithm for calculating the derivatives and the Cayley transform for preserving the projection orthogonality. Through simulation study under six different scenarios, we compare the proposed method to several benchmarks including least absolute shrinkage and selection operator, support vector machine, random forest, extreme learning machine, and multi-layer perceptron. It is shown that the proposed xNN model keeps the flexibility of pursuing high prediction accuracy while attaining improved interpretability. Finally, a real data example is employed as a showcase application.