Neuro-symbolic artificial intelligence (AI) excels at learning from noisy and generalized patterns, conducting logical inferences, and providing interpretable reasoning. Comprising a 'neuro' component for feature extraction and a 'symbolic' component for decision-making, neuro-symbolic AI has yet to fully benefit from efficient hardware accelerators. Additionally, current hardware struggles to accommodate applications requiring dynamic resource allocation between these two components. To address these challenges-and mitigate the typical data-transfer bottleneck of classical Von Neumann architectures-we propose a ferroelectric charge-domain compute-in-memory (CiM) array as the foundational processing element for neuro-symbolic AI. This array seamlessly handles both the critical multiply-accumulate (MAC) operations of the 'neuro' workload and the parallel associative search operations of the 'symbolic' workload. To enable this approach, we introduce an innovative 1FeFET-1C cell, combining a ferroelectric field-effect transistor (FeFET) with a capacitor. This design, overcomes the destructive sensing limitations of DRAM in CiM applications, while capable of capitalizing decades of DRAM expertise with a similar cell structure as DRAM, achieves high immunity against FeFET variation-crucial for neuro-symbolic AI-and demonstrates superior energy efficiency. The functionalities of our design have been successfully validated through SPICE simulations and prototype fabrication and testing. Our hardware platform has been benchmarked in executing typical neuro-symbolic AI reasoning tasks, showing over 2x improvement in latency and 1000x improvement in energy efficiency compared to GPU-based implementations.
We introduce CompAct, a technique that reduces peak memory utilization on GPU by 25-30% for pretraining and 50% for fine-tuning of LLMs. Peak device memory is a major limiting factor in training LLMs, with various recent works aiming to reduce model memory. However most works don't target the largest component of allocated memory during training: the model's compute graph, which is stored for the backward pass. By storing low-rank, compressed activations to be used in the backward pass we greatly reduce the required memory, unlike previous methods which only reduce optimizer overheads or the number of trained parameters. Our compression uses random projection matrices, thus avoiding additional memory overheads. Comparisons with previous techniques for either pretraining or fine-tuning show that CompAct substantially improves existing compute-performance tradeoffs. We expect CompAct's savings to scale even higher for larger models.
Nearly 900 million people live in low-lying coastal zones around the world and bear the brunt of impacts from more frequent and severe hurricanes and storm surges. Oceanographers simulate ocean current circulation along the coasts to develop early warning systems that save lives and prevent loss and damage to property from coastal hazards. Traditionally, such simulations are conducted using coastal ocean circulation models such as the Regional Ocean Modeling System (ROMS), which usually runs on an HPC cluster with multiple CPU cores. However, the process is time-consuming and energy expensive. While coarse-grained ROMS simulations offer faster alternatives, they sacrifice detail and accuracy, particularly in complex coastal environments. Recent advances in deep learning and GPU architecture have enabled the development of faster AI (neural network) surrogates. This paper introduces an AI surrogate based on a 4D Swin Transformer to simulate coastal tidal wave propagation in an estuary for both hindcast and forecast (up to 12 days). Our approach not only accelerates simulations but also incorporates a physics-based constraint to detect and correct inaccurate results, ensuring reliability while minimizing manual intervention. We develop a fully GPU-accelerated workflow, optimizing the model training and inference pipeline on NVIDIA DGX-2 A100 GPUs. Our experiments demonstrate that our AI surrogate reduces the time cost of 12-day forecasting of traditional ROMS simulations from 9,908 seconds (on 512 CPU cores) to 22 seconds (on one A100 GPU), achieving over 450$\times$ speedup while maintaining high-quality simulation results. This work contributes to oceanographic modeling by offering a fast, accurate, and physically consistent alternative to traditional simulation models, particularly for real-time forecasting in rapid disaster response.
Large language models (LLMs) with long context windows have gained significant attention. However, the KV cache, stored to avoid re-computation, becomes a bottleneck. Various dynamic sparse or TopK-based attention approximation methods have been proposed to leverage the common insight that attention is sparse. In this paper, we first show that TopK attention itself suffers from quality degradation in certain downstream tasks because attention is not always as sparse as expected. Rather than selecting the keys and values with the highest attention scores, sampling with theoretical guarantees can provide a better estimation for attention output. To make the sampling-based approximation practical in LLM generation, we propose MagicPIG, a heterogeneous system based on Locality Sensitive Hashing (LSH). MagicPIG significantly reduces the workload of attention computation while preserving high accuracy for diverse tasks. MagicPIG stores the LSH hash tables and runs the attention computation on the CPU, which allows it to serve longer contexts and larger batch sizes with high approximation accuracy. MagicPIG can improve decoding throughput by up to $5\times$ across various GPU hardware and achieve 54ms decoding latency on a single RTX 4090 for Llama-3.1-8B-Instruct model with a context of 96k tokens. The code is available at https://github.com/Infini-AI-Lab/MagicPIG.
Reconstructing cosmological initial conditions (ICs) from late-time observations is a difficult task, which relies on the use of computationally expensive simulators alongside sophisticated statistical methods to navigate multi-million dimensional parameter spaces. We present a simple method for Bayesian field reconstruction based on modeling the posterior distribution of the initial matter density field to be diagonal Gaussian in Fourier space, with its covariance and the mean estimator being the trainable parts of the algorithm. Training and sampling are extremely fast (training: $\sim 1 \, \mathrm{h}$ on a GPU, sampling: $\lesssim 3 \, \mathrm{s}$ for 1000 samples at resolution $128^3$), and our method supports industry-standard (non-differentiable) $N$-body simulators. We verify the fidelity of the obtained IC samples in terms of summary statistics.
Multimodal large language models (MLLMs) have made significant strides, yet they face challenges in the medical domain due to limited specialized knowledge. While recent medical MLLMs demonstrate strong performance in lab settings, they often struggle in real-world applications, highlighting a substantial gap between research and practice. In this paper, we seek to address this gap at various stages of the end-to-end learning pipeline, including data collection, model fine-tuning, and evaluation. At the data collection stage, we introduce SemiHVision, a dataset that combines human annotations with automated augmentation techniques to improve both medical knowledge representation and diagnostic reasoning. For model fine-tuning, we trained PMC-Cambrian-8B-AN over 2400 H100 GPU hours, resulting in performance that surpasses public medical models like HuatuoGPT-Vision-34B (79.0% vs. 66.7%) and private general models like Claude3-Opus (55.7%) on traditional benchmarks such as SLAKE and VQA-RAD. In the evaluation phase, we observed that traditional benchmarks cannot accurately reflect realistic clinical task capabilities. To overcome this limitation and provide more targeted guidance for model evaluation, we introduce the JAMA Clinical Challenge, a novel benchmark specifically designed to evaluate diagnostic reasoning. On this benchmark, PMC-Cambrian-AN achieves state-of-the-art performance with a GPT-4 score of 1.29, significantly outperforming HuatuoGPT-Vision-34B (1.13) and Claude3-Opus (1.17), demonstrating its superior diagnostic reasoning abilities.
Score identity Distillation (SiD) is a data-free method that has achieved SOTA performance in image generation by leveraging only a pretrained diffusion model, without requiring any training data. However, its ultimate performance is constrained by how accurate the pretrained model captures the true data scores at different stages of the diffusion process. In this paper, we introduce SiDA (SiD with Adversarial Loss), which not only enhances generation quality but also improves distillation efficiency by incorporating real images and adversarial loss. SiDA utilizes the encoder from the generator's score network as a discriminator, allowing it to distinguish between real images and those generated by SiD. The adversarial loss is batch-normalized within each GPU and then combined with the original SiD loss. This integration effectively incorporates the average "fakeness" per GPU batch into the pixel-based SiD loss, enabling SiDA to distill a single-step generator. SiDA converges significantly faster than its predecessor when distilled from scratch, and swiftly improves upon the original model's performance during fine-tuning from a pre-distilled SiD generator. This one-step adversarial distillation method establishes new benchmarks in generation performance when distilling EDM diffusion models, achieving FID scores of 1.110 on ImageNet 64x64. When distilling EDM2 models trained on ImageNet 512x512, our SiDA method surpasses even the largest teacher model, EDM2-XXL, which achieved an FID of 1.81 using classifier-free guidance (CFG) and 63 generation steps. In contrast, SiDA achieves FID scores of 2.156 for size XS, 1.669 for S, 1.488 for M, 1.413 for L, 1.379 for XL, and 1.366 for XXL, all without CFG and in a single generation step. These results highlight substantial improvements across all model sizes. Our code is available at https://github.com/mingyuanzhou/SiD/tree/sida.
3D Gaussian Splatting has shown fast and high-quality rendering results in static scenes by leveraging dense 3D prior and explicit representations. Unfortunately, the benefits of the prior and representation do not involve novel view synthesis for dynamic motions. Ironically, this is because the main barrier is the reliance on them, which requires increasing training and rendering times to account for dynamic motions. In this paper, we design a Explicit 4D Gaussian Splatting(Ex4DGS). Our key idea is to firstly separate static and dynamic Gaussians during training, and to explicitly sample positions and rotations of the dynamic Gaussians at sparse timestamps. The sampled positions and rotations are then interpolated to represent both spatially and temporally continuous motions of objects in dynamic scenes as well as reducing computational cost. Additionally, we introduce a progressive training scheme and a point-backtracking technique that improves Ex4DGS's convergence. We initially train Ex4DGS using short timestamps and progressively extend timestamps, which makes it work well with a few point clouds. The point-backtracking is used to quantify the cumulative error of each Gaussian over time, enabling the detection and removal of erroneous Gaussians in dynamic scenes. Comprehensive experiments on various scenes demonstrate the state-of-the-art rendering quality from our method, achieving fast rendering of 62 fps on a single 2080Ti GPU.
Democratization of AI is an important topic within the broader topic of the digital divide. This issue is relevant to LLMs, which are becoming popular as AI co-pilots but suffer from a lack of accessibility due to high computational demand. In this study, we evaluate whether quantization is a viable approach toward enabling LLMs on generic consumer devices. The study assesses the performance of five quantized code LLMs in Lua code generation tasks. To evaluate the impact of quantization, the models with 7B parameters were tested on a consumer laptop at 2-, 4-, and 8-bit integer precisions and compared to non-quantized code LLMs with 1.3, 2, and 3 billion parameters. Lua is chosen as a low-level resource language to avoid models' biases related to high-resource languages. The results suggest that the models quantized at the 4-bit integer precision offer the best trade-off between performance and model size. These models can be comfortably deployed on an average laptop without a dedicated GPU. The performance significantly drops at the 2-bit integer precision. The models at 8-bit integer precision require more inference time that does not effectively translate to better performance. The 4-bit models with 7 billion parameters also considerably outperform non-quantized models with lower parameter numbers despite having comparable model sizes with respect to storage and memory demand. While quantization indeed increases the accessibility of smaller LLMs with 7 billion parameters, these LLMs demonstrate overall low performance (less than 50\%) on high-precision and low-resource tasks such as Lua code generation. While accessibility is improved, usability is still not at the practical level comparable to foundational LLMs such as GPT-4o or Llama 3.1 405B.
Large language models (LLMs) have garnered substantial attention due to their promising applications in diverse domains. Nevertheless, the increasing size of LLMs comes with a significant surge in the computational requirements for training and deployment. Memristor crossbars have emerged as a promising solution, which demonstrated a small footprint and remarkably high energy efficiency in computer vision (CV) models. Memristors possess higher density compared to conventional memory technologies, making them highly suitable for effectively managing the extreme model size associated with LLMs. However, deploying LLMs on memristor crossbars faces three major challenges. Firstly, the size of LLMs increases rapidly, already surpassing the capabilities of state-of-the-art memristor chips. Secondly, LLMs often incorporate multi-head attention blocks, which involve non-weight stationary multiplications that traditional memristor crossbars cannot support. Third, while memristor crossbars excel at performing linear operations, they are not capable of executing complex nonlinear operations in LLM such as softmax and layer normalization. To address these challenges, we present a novel architecture for the memristor crossbar that enables the deployment of state-of-the-art LLM on a single chip or package, eliminating the energy and time inefficiencies associated with off-chip communication. Our testing on BERT_Large showed negligible accuracy loss. Compared to traditional memristor crossbars, our architecture achieves enhancements of up to 39X in area overhead and 18X in energy consumption. Compared to modern TPU/GPU systems, our architecture demonstrates at least a 68X reduction in the area-delay product and a significant 69% energy consumption reduction.
A persistence diagram provides a compact summary of persistent homology, which captures the topological features of a space at different scales. However, due to its nature as a set, incorporating it as a feature into a machine learning framework is challenging. Several methods have been proposed to use persistence diagrams as input for machine learning models, but they often require complex preprocessing steps and extensive hyperparameter tuning. In this paper, we propose a novel transformer architecture called the \textit{Extended Persistence Transformer (xPerT)}, which is highly scalable than the compared to Persformer, an existing transformer for persistence diagrams. xPerT reduces GPU memory usage by over 90\% and improves accuracy on multiple datasets. Additionally, xPerT does not require complex preprocessing steps or extensive hyperparameter tuning, making it easy to use in practice. Our code is available at https://github.com/sehunfromdaegu/xpert.
The inverse of an invertible convolution is an important operation that comes up in Normalizing Flows, Image Deblurring, etc. The naive algorithm for backpropagation of this operation using Gaussian elimination has running time $O(n^3)$ where $n$ is the number of pixels in the image. We give a fast parallel backpropagation algorithm with running time $O(\sqrt{n})$ for a square image and provide a GPU implementation of the same. Inverse of Convolutions are usually used in Normalizing Flows in the sampling pass, making them slow. We propose to use the Inverse of Convolutions in the forward (image to latent vector) pass of the Normalizing flow. Since the sampling pass is the inverse of the forward pass, it will use convolutions only, resulting in efficient sampling times. We use our parallel backpropagation algorithm to optimize the inverse of the convolution layer, resulting in fast training times. We implement this approach in various Normalizing Flow backbones, resulting in our Inverse-Flow models. We benchmark Inverse-Flow on standard datasets and show significantly improved sampling times with similar bits per dimension compared to previous models.
Current and future large scale structure surveys aim to constrain the neutrino mass and the equation of state of dark energy. We aim to construct accurate and interpretable symbolic approximations to the linear and nonlinear matter power spectra as a function of cosmological parameters in extended $\Lambda$CDM models which contain massive neutrinos and non-constant equations of state for dark energy. This constitutes an extension of the syren-halofit emulators to incorporate these two effects, which we call syren-new (SYmbolic-Regression-ENhanced power spectrum emulator with NEutrinos and $W_0-w_a$). We also obtain a simple approximation to the derived parameter $\sigma_8$ as a function of the cosmological parameters for these models. Our results for the linear power spectrum are designed to emulate CLASS, whereas for the nonlinear case we aim to match the results of EuclidEmulator2. We compare our results to existing emulators and $N$-body simulations. Our analytic emulators for $\sigma_8$, the linear and nonlinear power spectra achieve root mean squared errors of 0.1%, 0.3% and 1.3%, respectively, across a wide range of cosmological parameters, redshifts and wavenumbers. We verify that emulator-related discrepancies are subdominant compared to observational errors and other modelling uncertainties when computing shear power spectra for LSST-like surveys. Our expressions have similar accuracy to existing (numerical) emulators, but are at least an order of magnitude faster, both on a CPU and GPU. Our work greatly improves the accuracy, speed and range of applicability of current symbolic approximations to the linear and nonlinear matter power spectra. We provide publicly available code for all symbolic approximations found.
Existing audio-driven facial animation methods face critical challenges, including expression leakage, ineffective subtle expression transfer, and imprecise audio-driven synchronization. We discovered that these issues stem from limitations in motion representation and the lack of fine-grained control over facial expressions. To address these problems, we present Takin-ADA, a novel two-stage approach for real-time audio-driven portrait animation. In the first stage, we introduce a specialized loss function that enhances subtle expression transfer while reducing unwanted expression leakage. The second stage utilizes an advanced audio processing technique to improve lip-sync accuracy. Our method not only generates precise lip movements but also allows flexible control over facial expressions and head motions. Takin-ADA achieves high-resolution (512x512) facial animations at up to 42 FPS on an RTX 4090 GPU, outperforming existing commercial solutions. Extensive experiments demonstrate that our model significantly surpasses previous methods in video quality, facial dynamics realism, and natural head movements, setting a new benchmark in the field of audio-driven facial animation.
Recent advancements in latent diffusion models (LDMs) have markedly enhanced text-to-audio generation, yet their iterative sampling processes impose substantial computational demands, limiting practical deployment. While recent methods utilizing consistency-based distillation aim to achieve few-step or single-step inference, their one-step performance is constrained by curved trajectories, preventing them from surpassing traditional diffusion models. In this work, we introduce FlashAudio with rectified flows to learn straight flow for fast simulation. To alleviate the inefficient timesteps allocation and suboptimal distribution of noise, FlashAudio optimizes the time distribution of rectified flow with Bifocal Samplers and proposes immiscible flow to minimize the total distance of data-noise pairs in a batch vias assignment. Furthermore, to address the amplified accumulation error caused by the classifier-free guidance (CFG), we propose Anchored Optimization, which refines the guidance scale by anchoring it to a reference trajectory. Experimental results on text-to-audio generation demonstrate that FlashAudio's one-step generation performance surpasses the diffusion-based models with hundreds of sampling steps on audio quality and enables a sampling speed of 400x faster than real-time on a single NVIDIA 4090Ti GPU. Code will be available at https://github.com/liuhuadai/FlashAudio.
Energy consumption has become a critical design metric and a limiting factor in the development of future computing architectures, from small wearable devices to large-scale leadership computing facilities. The predominant methods in energy management optimization are focused on CPUs. However, GPUs are increasingly significant and account for the majority of energy consumption in heterogeneous high performance computing (HPC) systems. Moreover, they typically rely on either purely offline training or a hybrid of offline and online training, which are impractical and lead to energy loss during data collection. Therefore, this paper studies a novel and practical online energy optimization problem for GPUs in HPC scenarios. The problem is challenging due to the inherent performance-energy trade-offs of GPUs, the exploration & exploitation dilemma across frequencies, and the lack of explicit performance counters in GPUs. To address these challenges, we formulate the online energy consumption optimization problem as a multi-armed bandit framework and develop a novel bandit based framework EnergyUCB. EnergyUCB is designed to dynamically adjust GPU core frequencies in real-time, reducing energy consumption with minimal impact on performance. Specifically, the proposed framework EnergyUCB (1) balances the performance-energy trade-off in the reward function, (2) effectively navigates the exploration & exploitation dilemma when adjusting GPU core frequencies online, and (3) leverages the ratio of GPU core utilization to uncore utilization as a real-time GPU performance metric. Experiments on a wide range of real-world HPC benchmarks demonstrate that EnergyUCB can achieve substantial energy savings. The code of EnergyUCB is available at https://github.com/XiongxiaoXu/EnergyUCB-Bandit.
Time series forecasting is vital in numerous web applications, influencing critical decision-making across industries. While diffusion models have recently gained increasing popularity for this task, we argue they suffer from a significant drawback: indiscriminate noise addition to the original time series followed by denoising, which can obscure underlying dynamic evolving trend and complicate forecasting. To address this limitation, we propose a novel flexible decoupled framework (FDF) that learns high-quality time series representations for enhanced forecasting performance. A key characteristic of our approach leverages the inherent inductive bias of time series data of its decomposed trend and seasonal components, each modeled separately to enable decoupled analysis and modeling. Specifically, we propose an innovative Conditional Denoising Seasonal Module (CDSM) within the diffusion model, which leverages statistical information from the historical window to conditionally model the complex seasonal component. Notably, we incorporate a Polynomial Trend Module (PTM) to effectively capture the smooth trend component, thereby enhancing the model's ability to represent temporal dependencies. Extensive experiments validate the effectiveness of our framework, demonstrating superior performance over existing methods and highlighting its flexibility in time series forecasting. The source code is available at https://github.com/zjt-gpu/FDF.
This work proposes a novel approach beyond supervised learning for effective pathological image analysis, addressing the challenge of limited robust labeled data. Pathological diagnosis of diseases like cancer has conventionally relied on the evaluation of morphological features by physicians and pathologists. However, recent advancements in compute-aided diagnosis (CAD) systems are gaining significant attention as diagnostic support tools. Although the advancement of deep learning has improved CAD significantly, segmentation models typically require large pixel-level annotated dataset, and such labeling is expensive. Existing studies not based on supervised approaches still struggle with limited generalization, and no practical approach has emerged yet. To address this issue, we present a weakly supervised semantic segmentation (WSSS) model by combining class activation map and Segment Anything Model (SAM)-based pseudo-labeling. For effective pretraining, we adopt the SAM-a foundation model that is pretrained on large datasets and operates in zero-shot configurations using only coarse prompts. The proposed approach transfer enhanced Attention Dropout Layer's knowledge to SAM, thereby generating pseudo-labels. To demonstrate the superiority of the proposed method, experimental studies are conducted on histopathological breast cancer datasets. The proposed method outperformed other WSSS methods across three datasets, demonstrating its efficiency by achieving this with only 12GB of GPU memory during training. Our code is available at : https://github.com/QI-NemoSong/EP-SAM
4D Gaussian Splatting (4DGS) has recently emerged as a promising technique for capturing complex dynamic 3D scenes with high fidelity. It utilizes a 4D Gaussian representation and a GPU-friendly rasterizer, enabling rapid rendering speeds. Despite its advantages, 4DGS faces significant challenges, notably the requirement of millions of 4D Gaussians, each with extensive associated attributes, leading to substantial memory and storage cost. This paper introduces a memory-efficient framework for 4DGS. We streamline the color attribute by decomposing it into a per-Gaussian direct color component with only 3 parameters and a shared lightweight alternating current color predictor. This approach eliminates the need for spherical harmonics coefficients, which typically involve up to 144 parameters in classic 4DGS, thereby creating a memory-efficient 4D Gaussian representation. Furthermore, we introduce an entropy-constrained Gaussian deformation technique that uses a deformation field to expand the action range of each Gaussian and integrates an opacity-based entropy loss to limit the number of Gaussians, thus forcing our model to use as few Gaussians as possible to fit a dynamic scene well. With simple half-precision storage and zip compression, our framework achieves a storage reduction by approximately 190$\times$ and 125$\times$ on the Technicolor and Neural 3D Video datasets, respectively, compared to the original 4DGS. Meanwhile, it maintains comparable rendering speeds and scene representation quality, setting a new standard in the field.
Reasoning, the ability to logically draw conclusions from existing knowledge, is a hallmark of human. Together with perception, they constitute the two major themes of artificial intelligence. While deep learning has pushed the limit of perception beyond human-level performance, the progress in reasoning domains is way behind. One fundamental reason is that reasoning problems usually have flexible structures for both knowledge and queries, and many existing models only perform well on structures seen during training. Here we aim to push the boundary of reasoning models by devising algorithms that generalize across knowledge and query structures, as well as systems that accelerate development on structured data. This thesis consists of three parts. In Part I, we study models that can inductively generalize to unseen knowledge graphs with new entity and relation vocabularies. For new entities, we propose a framework that learns neural operators in a dynamic programming algorithm computing path representations. For relations, we construct a relation graph to capture the interactions between relations, thereby converting new relations into new entities. In Part II, we propose two solutions for generalizing across multi-step queries on knowledge graphs and text respectively. For knowledge graphs, we show that multi-step queries can be solved by multiple calls of graph neural networks and fuzzy logic operations. For text, we devise an algorithm to learn explicit knowledge as textual rules to improve large language models on multi-step queries. In Part III, we propose two systems to facilitate machine learning development on structured data. Our library treats structured data as first-class citizens and removes the barrier for developing algorithms on structured data. Our node embedding system solves the GPU memory bottleneck of embedding matrices and scales to graphs with billion nodes.