Tajana Rosing

Md Mizanur Rahaman Nayan, Tianqi Zhang, Flavio Ponzina et al.

Vector database search with frequent updates is increasingly critical in applications such as retrieval augmented generation, recommendation systems, and large-scale embedding retrieval. Existing solutions, such as graph-based and partition-based approximate nearest neighbor search (ANNS), suffer from frequent index rebuilding due to data distribution-dependent indexing that impacts continuous deployment and causes long rebuilding latency. This paper proposes an algorithm-hardware co-designed platform, ACRONYM, that addresses key problems with state of the art database search. Algorithmically, it leverages efficient encoding independent of data distribution and Hamming-distance based search for efficient hardware acceleration. Architecturally, we propose CAM-based in-memory parallel distance computation followed by time multiplexed approximated top-k selection to enable the exhaustive search. We propose two-stage search that includes coarse search followed by binary refinement to achieve high recall in CAM based search which is heavily limited to small vector dimension due to capacity and wordline parasitic. ACRONYM supports continuous update without stalling and integrates novel XOR-and-Accumulate (XAC) based systolic-array encoder for efficient on chip encoding during search. Across million-scale datasets, while serving dynamic database ACRONYM achieves >90% recall at a throughput of 8e6 queries per second, with a memory footprint of only 32MB and an average energy consumption of 2.56uJ per query, speedup over HNSW (CPU) of about 400x and FAISS-IVF (GPU) of about 80x.

Keming Fan, Ashkan Moradifirouzabadi, Xiangjin Wu et al.

Mass spectrometry (MS) is essential for proteomics and metabolomics but faces impending challenges in efficiently processing the vast volumes of data. This paper introduces SpecPCM, an in-memory computing (IMC) accelerator designed to achieve substantial improvements in energy and delay efficiency for both MS spectral clustering and database (DB) search. SpecPCM employs analog processing with low-voltage swing and utilizes recently introduced phase change memory (PCM) devices based on superlattice materials, optimized for low-voltage and low-power programming. Our approach integrates contributions across multiple levels: application, algorithm, circuit, device, and instruction sets. We leverage a robust hyperdimensional computing (HD) algorithm with a novel dimension-packing method and develop specialized hardware for the end-to-end MS pipeline to overcome the non-ideal behavior of PCM devices. We further optimize multi-level PCM devices for different tasks by using different materials. We also perform a comprehensive design exploration to improve energy and delay efficiency while maintaining accuracy, exploring various combinations of hardware and software parameters controlled by the instruction set architecture (ISA). SpecPCM, with up to three bits per cell, achieves speedups of up to 82x and 143x for MS clustering and DB search tasks, respectively, along with a four-orders-of-magnitude improvement in energy efficiency compared with state-of-the-art CPU/GPU tools.

Xiaofan Yu, Ludmila Cherkasova, Harsh Vardhan et al.

Federated Learning (FL) has gained increasing interest in recent years as a distributed on-device learning paradigm. However, multiple challenges remain to be addressed for deploying FL in real-world Internet-of-Things (IoT) networks with hierarchies. Although existing works have proposed various approaches to account data heterogeneity, system heterogeneity, unexpected stragglers and scalibility, none of them provides a systematic solution to address all of the challenges in a hierarchical and unreliable IoT network. In this paper, we propose an asynchronous and hierarchical framework (Async-HFL) for performing FL in a common three-tier IoT network architecture. In response to the largely varied delays, Async-HFL employs asynchronous aggregations at both the gateway and the cloud levels thus avoids long waiting time. To fully unleash the potential of Async-HFL in converging speed under system heterogeneities and stragglers, we design device selection at the gateway level and device-gateway association at the cloud level. Device selection chooses edge devices to trigger local training in real-time while device-gateway association determines the network topology periodically after several cloud epochs, both satisfying bandwidth limitation. We evaluate Async-HFL's convergence speedup using large-scale simulations based on ns-3 and a network topology from NYCMesh. Our results show that Async-HFL converges 1.08-1.31x faster in wall-clock time and saves up to 21.6% total communication cost compared to state-of-the-art asynchronous FL algorithms (with client selection). We further validate Async-HFL on a physical deployment and observe robust convergence under unexpected stragglers.

Xiaofan Yu, Yunhui Guo, Sicun Gao et al.

Unsupervised lifelong learning refers to the ability to learn over time while memorizing previous patterns without supervision. Although great progress has been made in this direction, existing work often assumes strong prior knowledge about the incoming data (e.g., knowing the class boundaries), which can be impossible to obtain in complex and unpredictable environments. In this paper, motivated by real-world scenarios, we propose a more practical problem setting called online self-supervised lifelong learning without prior knowledge. The proposed setting is challenging due to the non-iid and single-pass data, the absence of external supervision, and no prior knowledge. To address the challenges, we propose Self-Supervised ContrAstive Lifelong LEarning without Prior Knowledge (SCALE) which can extract and memorize representations on the fly purely from the data continuum. SCALE is designed around three major components: a pseudo-supervised contrastive loss, a self-supervised forgetting loss, and an online memory update for uniform subset selection. All three components are designed to work collaboratively to maximize learning performance. We perform comprehensive experiments of SCALE under iid and four non-iid data streams. The results show that SCALE outperforms the state-of-the-art algorithm in all settings with improvements up to 3.83%, 2.77% and 5.86% in terms of kNN accuracy on CIFAR-10, CIFAR-100, and TinyImageNet datasets.

Lanxiang Hu, Siqi Kou, Yichao Fu et al.

Multi-token generation has emerged as a promising paradigm for accelerating transformer-based large model inference. Recent efforts primarily explore diffusion Large Language Models (dLLMs) for parallel decoding to reduce inference latency. To achieve AR-level generation quality, many techniques adapt AR models into dLLMs to enable parallel decoding. However, they suffer from limited speedup compared to AR models due to a pretrain-to-posttrain mismatch. Specifically, the masked data distribution in post-training deviates significantly from the real-world data distribution seen during pretraining, and dLLMs rely on bidirectional attention, which conflicts with the causal prior learned during pretraining and hinders the integration of exact KV cache reuse. To address this, we introduce Jacobi Forcing, a progressive distillation paradigm where models are trained on their own generated parallel decoding trajectories, smoothly shifting AR models into efficient parallel decoders while preserving their pretrained causal inference property. The models trained under this paradigm, Jacobi Forcing Model, achieves 3.8x wall-clock speedup on coding and math benchmarks with minimal loss in performance. Based on Jacobi Forcing Models' trajectory characteristics, we introduce multi-block decoding with rejection recycling, which enables up to 4.5x higher token acceptance count per iteration and nearly 4.0x wall-clock speedup, effectively trading additional compute for lower inference latency. Our code is available at https://github.com/hao-ai-lab/JacobiForcing.

Onat Gungor, Tajana Rosing, Baris Aksanli

Industrial Internet of Things (I-IoT) enables fully automated production systems by continuously monitoring devices and analyzing collected data. Machine learning methods are commonly utilized for data analytics in such systems. Cyber-attacks are a grave threat to I-IoT as they can manipulate legitimate inputs, corrupting ML predictions and causing disruptions in the production systems. Hyper-dimensional computing (HDC) is a brain-inspired machine learning method that has been shown to be sufficiently accurate while being extremely robust, fast, and energy-efficient. In this work, we use HDC for intelligent fault diagnosis against different adversarial attacks. Our black-box adversarial attacks first train a substitute model and create perturbed test instances using this trained model. These examples are then transferred to the target models. The change in the classification accuracy is measured as the difference before and after the attacks. This change measures the resiliency of a learning method. Our experiments show that HDC leads to a more resilient and lightweight learning solution than the state-of-the-art deep learning methods. HDC has up to 67.5% higher resiliency compared to the state-of-the-art methods while being up to 25.1% faster to train.

Anthony Thomas, Behnam Khaleghi, Gopi Krishna Jha et al.

Hyperdimensional computing (HDC) is a paradigm for data representation and learning originating in computational neuroscience. HDC represents data as high-dimensional, low-precision vectors which can be used for a variety of information processing tasks like learning or recall. The mapping to high-dimensional space is a fundamental problem in HDC, and existing methods encounter scalability issues when the input data itself is high-dimensional. In this work, we explore a family of streaming encoding techniques based on hashing. We show formally that these methods enjoy comparable guarantees on performance for learning applications while being substantially more efficient than existing alternatives. We validate these results experimentally on a popular high-dimensional classification problem and show that our approach easily scales to very large data sets.

Jason Kong, Nilesh Prasad Pandey, Flavio Ponzina et al.

Deploying Large Language Models (LLMs) on edge devices faces severe computational and memory constraints, limiting real-time processing and on-device intelligence. Hybrid architectures combining Structured State Space Models (SSMs) with transformer-based LLMs offer a balance of efficiency and performance. Aggressive quantization can drastically cut model size and speed up inference, but its uneven effects on different components require careful management. In this work, we propose a lightweight, backpropagation-free, surrogate-based sensitivity analysis framework to identify hybrid SSM-Transformer components most susceptible to quantization-induced degradation. Relying solely on forward-pass metrics, our method avoids expensive gradient computations and retraining, making it suitable for situations where access to in-domain data is limited due to proprietary restrictions or privacy constraints. We also provide a formal analysis showing that the Kullback-Leibler (KL) divergence metric better captures quantization sensitivity for Language modeling tasks than widely adopted alternatives such as mean squared error (MSE) and signal-to-quantization-noise ratio (SQNR). Through extensive experiments on SSM and hybrid architectures, our ablation studies confirm that KL-based rankings align with observed performance drops and outperform alternative metrics. This framework enables the practical deployment of advanced hybrid models on resource-constrained edge devices with minimal accuracy loss. We further validate our approach with real-world on-device profiling on Intel Lunar Lake hardware, demonstrating that KL-guided mixed-precision achieves near-FP16 perplexity with model sizes and throughput competitive with Uniform INT4 on both CPU and GPU execution modes. Code is available at https://github.com/jasonkongie/kl-ssm-quant.

Onat Gungor, Tajana Rosing, Baris Aksanli

Industrial Internet of Things (I-IoT) is a collaboration of devices, sensors, and networking equipment to monitor and collect data from industrial operations. Machine learning (ML) methods use this data to make high-level decisions with minimal human intervention. Data-driven predictive maintenance (PDM) is a crucial ML-based I-IoT application to find an optimal maintenance schedule for industrial assets. The performance of these ML methods can seriously be threatened by adversarial attacks where an adversary crafts perturbed data and sends it to the ML model to deteriorate its prediction performance. The models should be able to stay robust against these attacks where robustness is measured by how much perturbation in input data affects model performance. Hence, there is a need for effective defense mechanisms that can protect these models against adversarial attacks. In this work, we propose a double defense mechanism to detect and mitigate adversarial attacks in I-IoT environments. We first detect if there is an adversarial attack on a given sample using novelty detection algorithms. Then, based on the outcome of our algorithm, marking an instance as attack or normal, we select adversarial retraining or standard training to provide a secondary defense layer. If there is an attack, adversarial retraining provides a more robust model, while we apply standard training for regular samples. Since we may not know if an attack will take place, our adaptive mechanism allows us to consider irregular changes in data. The results show that our double defense strategy is highly efficient where we can improve model robustness by up to 64.6% and 52% compared to standard and adversarial retraining, respectively.

Jing Chen, Abhijay Deevi, Onat Gungor et al.

The Controller Area Network (CAN) is a safety-critical in-vehicle communication protocol that lacks built-in security mechanisms, making intrusion detection essential. Existing approaches predominantly formulate CAN intrusion detection as a classification task, mapping complex traffic patterns to attack labels. However, this formulation abstracts away the temporal and relational structure of CAN traffic and misaligns with real-world forensic workflows, which require systematic reasoning about traffic behavior. To address this gap, we introduce CAN-QA, the first benchmark that reformulates CAN traffic analysis as a question-answering (QA) task. CAN-QA converts raw CAN logs into temporally segmented windows and applies deterministic rule-based templates to generate natural-language questions paired with automatically derived ground-truth answers. The resulting dataset comprises 33,128 QA pairs across 10 categories, each targeting distinct semantic and temporal properties of CAN traffic. Using CAN-QA, we evaluate large language models across both True/False and multiple-choice formats. Our results indicate that, although these models capture superficial statistical regularities, they struggle with temporal reasoning, multi-condition inference, and higher-level behavioral interpretation. Our code is available at https://github.com/Kriiiiss/CAN-QA.

Sumukh Pinge, Weihong Xu, Jaeyoung Kang et al.

Mass spectrometry-based proteomics is a key enabler for personalized healthcare, providing a deep dive into the complex protein compositions of biological systems. This technology has vast applications in biotechnology and biomedicine but faces significant computational bottlenecks. Current methodologies often require multiple hours or even days to process extensive datasets, particularly in the domain of spectral clustering. To tackle these inefficiencies, we introduce SpecHD, a hyperdimensional computing (HDC) framework supplemented by an FPGA-accelerated architecture with integrated near-storage preprocessing. Utilizing streamlined binary operations in an HDC environment, SpecHD capitalizes on the low-latency and parallel capabilities of FPGAs. This approach markedly improves clustering speed and efficiency, serving as a catalyst for real-time, high-throughput data analysis in future healthcare applications. Our evaluations demonstrate that SpecHD not only maintains but often surpasses existing clustering quality metrics while drastically cutting computational time. Specifically, it can cluster a large-scale human proteome dataset-comprising 25 million MS/MS spectra and 131 GB of MS data-in just 5 minutes. With energy efficiency exceeding 31x and a speedup factor that spans a range of 6x to 54x over existing state of-the-art solutions, SpecHD emerges as a promising solution for the rapid analysis of mass spectrometry data with great implications for personalized healthcare.

Lanxiang Hu, Abhilash Shankarampeta, Yixin Huang et al.

The next frontier for video generation lies in developing models capable of zero-shot reasoning, where understanding real-world scientific laws is crucial for accurate physical outcome modeling under diverse conditions. However, existing video benchmarks are physical commonsense-based, offering limited insight into video models' scientific reasoning capability. We introduce VideoScience-Bench, a benchmark designed to evaluate undergraduate-level scientific understanding in video models. Each prompt encodes a composite scientific scenario that requires understanding and reasoning across multiple scientific concepts to generate the correct phenomenon. The benchmark comprises 200 carefully curated prompts spanning 14 topics and 103 concepts in physics and chemistry. We conduct expert-annotated evaluations across seven state-of-the-art video models in T2V and I2V settings along five dimensions: Prompt Consistency, Phenomenon Congruency, Correct Dynamism, Immutability, and Spatio-Temporal Continuity. Using a VLM-as-a-Judge to assess video generations, we observe strong correlation with human assessments. To the best of our knowledge, VideoScience-Bench is the first benchmark to evaluate video models not only as generators but also as reasoners, requiring their generations to demonstrate scientific understanding consistent with expected physical and chemical phenomena. Our data and evaluation code are available at: \href{https://github.com/hao-ai-lab/VideoScience}{github.com/hao-ai-lab/VideoScience}.

Haichao Yang, Chang Eun Song, Weihong Xu et al.

This paper introduces FSL-HDnn, an energy-efficient accelerator that implements the end-to-end pipeline of feature extraction, classification, and on-chip few-shot learning (FSL) through gradient-free learning techniques in a 40 nm CMOS process. At its core, FSL-HDnn integrates two low-power modules: Weight clustering feature extractor and Hyperdimensional Computing (HDC). Feature extractor utilizes advanced weight clustering and pattern reuse strategies for optimized CNN-based feature extraction. Meanwhile, HDC emerges as a novel approach for lightweight FSL classifier, employing hyperdimensional vectors to improve training accuracy significantly compared to traditional distance-based approaches. This dual-module synergy not only simplifies the learning process by eliminating the need for complex gradients but also dramatically enhances energy efficiency and performance. Specifically, FSL-HDnn achieves an Intensity unprecedented energy efficiency of 5.7 TOPS/W for feature 1 extraction and 0.78 TOPS/W for classification and learning Training Intensity phases, achieving improvements of 2.6X and 6.6X, respectively, Storage over current state-of-the-art CNN and FSL processors.

Gamze Kirman Tokgoz, Onat Gungor, Tajana Rosing et al.

Time-series forecasting aims to predict future values by modeling temporal dependencies in historical observations. It is a critical component of many real-world systems, where accurate forecasts improve operational efficiency and help mitigate uncertainty and risk. More recently, machine learning (ML), and especially deep learning (DL)-based models, have gained widespread adoption for time-series forecasting, but they remain vulnerable to adversarial attacks. However, many state-of-the-art attack methods are not directly applicable in time-series settings, where storing complete historical data or performing attacks at every time step is often impractical. This paper proposes an adversarial attack framework for time-series forecasting under an online bounded-buffer setting, leveraging an informed and selective attack strategy. By selectively targeting time steps where the model exhibits high confidence and the expected prediction error is maximal, our framework produces fewer but substantially more effective attacks. Experiments show that our framework can increase the prediction error up to 2.42x, while performing attacks in fewer than 10% of time steps.

Matilda Gaddi, Jin Noh, Onat Gungor et al.

Large language models (LLMs) are increasingly applied to cybersecurity question answering (QA) for critical tasks such as incident response and vulnerability analysis. However, real-world operational contexts, including system logs and network configurations, inherently contain sensitive identifiers, e.g., IP addresses, host names, and user accounts. Processing this data with cloud-based models is often unsafe or infeasible in regulated environments. Furthermore, progress in privacy-preserving QA is hindered by the lack of annotated, context-rich datasets capable of jointly evaluating operational reasoning and privacy preservation. To address this gap, we introduce CYBERMASKQA, a privacy-aware QA benchmark covering key security domains. Unlike existing benchmarks that primarily test factual knowledge, CYBERMASKQA grounds questions in realistic organizational contexts with explicit causal dependencies among assets and privileges. Generated through a systematic pipeline, the dataset combines human-curated base scenarios with LLM-driven semantic expansion, annotating each instance with precise private entity labels to enable controlled information disclosure. Evaluations of QA accuracy and masking performance demonstrate the benchmark's utility for developing deployable, context-aware cybersecurity models and facilitating nuanced studies of privacy-utility trade-offs. Upon acceptance, we will release the dataset and the generation framework.

Weihong Xu, Junwei Chen, Po-Kai Hsu et al.

Approximate nearest neighbor search (ANNS) plays an indispensable role in a wide variety of applications, including recommendation systems, information retrieval, and semantic search. Among the cutting-edge ANNS algorithms, graph-based approaches provide superior accuracy and scalability on massive datasets. However, the best-performing graph-based ANN search solutions incur tens of hundreds of memory footprints as well as costly distance computation, thus hindering their efficient deployment at scale. The 3D NAND flash is emerging as a promising device for data-intensive applications due to its high density and nonvolatility. In this work, we present the near-storage processing (NSP)-based ANNS solution Proxima, to accelerate graph-based ANNS with algorithm-hardware co-design in 3D NAND flash. Proxima significantly reduces the complexity of graph search by leveraging the distance approximation and early termination. On top of the algorithmic enhancement, we implement Proxima search algorithm in 3D NAND flash using the heterogeneous integration technique. To maximize 3D NAND's bandwidth utilization, we present customized dataflow and optimized data allocation scheme. Our evaluation results show that: compared to graph ANNS on CPU and GPU, Proxima achieves a magnitude improvement in throughput or energy efficiency. Proxima yields 7x to 13x speedup over existing ASIC designs. Furthermore, Proxima achieves a good balance between accuracy, efficiency and storage density compared to previous NSP-based accelerators.

Yichao Fu, Junda Chen, Siqi Zhu et al.

Test-time reasoning algorithms such as chain-of-thought, self-consistency, and MCTS enhance LLM problem-solving but can wastefully generate many tokens without improving accuracy. At the same time, we observe that these algorithms exhibit answer stabilization: their intermediate solutions often cease to change after a certain point, and further investment of compute does not change their final answer. To quantify this phenomenon, we introduce Certaindex, an algorithm-agnostic metric measuring this evolving stability, signaling when further computation is unlikely to alter the final result. Certaindex is lightweight, can accelerate reasoning program inference via early exit, and further enables dynamic token allocation, gang scheduling, and many opportunities when integrated with real-world LLM serving systems. To quantify real-world benefits, we built Certaindex as a scheduler into Dynasor, our reasoning-aware LLM serving system, and demonstrate up to 50% compute savings and 3.3x higher throughput in real workloads with no accuracy drop. Our code is available at https://github.com/hao-ai-lab/Dynasor.git

Le Zhang, Onat Gungor, Flavio Ponzina et al.

Ensemble learning is a meta-learning approach that combines the predictions of multiple learners, demonstrating improved accuracy and robustness. Nevertheless, ensembling models like Convolutional Neural Networks (CNNs) result in high memory and computing overhead, preventing their deployment in embedded systems. These devices are usually equipped with small batteries that provide power supply and might include energy-harvesting modules that extract energy from the environment. In this work, we propose E-QUARTIC, a novel Energy Efficient Edge Ensembling framework to build ensembles of CNNs targeting Artificial Intelligence (AI)-based embedded systems. Our design outperforms single-instance CNN baselines and state-of-the-art edge AI solutions, improving accuracy and adapting to varying energy conditions while maintaining similar memory requirements. Then, we leverage the multi-CNN structure of the designed ensemble to implement an energy-aware model selection policy in energy-harvesting AI systems. We show that our solution outperforms the state-of-the-art by reducing system failure rate by up to 40% while ensuring higher average output qualities. Ultimately, we show that the proposed design enables concurrent on-device training and high-quality inference execution at the edge, limiting the performance and energy overheads to less than 0.04%.

Lanxiang Hu, Mingjia Huo, Yuxuan Zhang et al.

Playing video games requires perception, memory, and planning, exactly the faculties modern large language model (LLM) agents are expected to master. We study the major challenges in using popular video games to evaluate modern LLMs and find that directly dropping LLMs into games cannot make an effective evaluation, for three reasons -- brittle vision perception, prompt sensitivity, and potential data contamination. We introduce lmgame-Bench to turn games into reliable evaluations. lmgame-Bench features a suite of platformer, puzzle, and narrative games delivered through a unified Gym-style API and paired with lightweight perception and memory scaffolds, and is designed to stabilize prompt variance and remove contamination. Across 13 leading models, we show lmgame-Bench is challenging while still separating models well. Correlation analysis shows that every game probes a unique blend of capabilities often tested in isolation elsewhere. More interestingly, performing reinforcement learning on a single game from lmgame-Bench transfers both to unseen games and to external planning tasks. Our evaluation code is available at https://github.com/lmgame-org/GamingAgent/lmgame-bench.

Xiaofan Yu, Anthony Thomas, Ivannia Gomez Moreno et al.

On-device learning has emerged as a prevailing trend that avoids the slow response time and costly communication of cloud-based learning. The ability to learn continuously and indefinitely in a changing environment, and with resource constraints, is critical for real sensor deployments. However, existing designs are inadequate for practical scenarios with (i) streaming data input, (ii) lack of supervision and (iii) limited on-board resources. In this paper, we design and deploy the first on-device lifelong learning system called LifeHD for general IoT applications with limited supervision. LifeHD is designed based on a novel neurally-inspired and lightweight learning paradigm called Hyperdimensional Computing (HDC). We utilize a two-tier associative memory organization to intelligently store and manage high-dimensional, low-precision vectors, which represent the historical patterns as cluster centroids. We additionally propose two variants of LifeHD to cope with scarce labeled inputs and power constraints. We implement LifeHD on off-the-shelf edge platforms and perform extensive evaluations across three scenarios. Our measurements show that LifeHD improves the unsupervised clustering accuracy by up to 74.8% compared to the state-of-the-art NN-based unsupervised lifelong learning baselines with as much as 34.3x better energy efficiency. Our code is available at https://github.com/Orienfish/LifeHD.

Ivannia Gomez Moreno, Yi Yao, Ye Tian et al.

LiDAR semantic segmentation plays a pivotal role in 3D scene understanding for edge applications such as autonomous driving. However, significant challenges remain for real-world deployments, particularly for on-device post-deployment adaptation. Real-world environments can shift as the system navigates through different locations, leading to substantial performance degradation without effective and timely model adaptation. Furthermore, edge systems operate under strict computational and energy constraints, making it infeasible to adapt conventional segmentation models (based on large neural networks) directly on-device. To address the above challenges, we introduce HyperLiDAR, the first lightweight, post-deployment LiDAR segmentation framework based on Hyperdimensional Computing (HDC). The design of HyperLiDAR fully leverages the fast learning and high efficiency of HDC, inspired by how the human brain processes information. To further improve the adaptation efficiency, we identify the high data volume per scan as a key bottleneck and introduce a buffer selection strategy that focuses learning on the most informative points. We conduct extensive evaluations on two state-of-the-art LiDAR segmentation benchmarks and two representative devices. Our results show that HyperLiDAR outperforms or achieves comparable adaptation performance to state-of-the-art segmentation methods, while achieving up to a 13.8x speedup in retraining.

Nilesh Prasad Pandey, Jangseon Park, Onat Gungor et al.

Deploying Small Language Models (SLMs) on edge platforms is critical for real-time, privacy-sensitive generative AI, yet constrained by memory, latency, and energy budgets. Quantization reduces model size and cost but suffers from device noise in emerging non-volatile memories, while conventional memory hierarchies further limit efficiency. SRAM provides fast access but has low density, DRAM must simultaneously accommodate static weights and dynamic KV caches, which creates bandwidth contention, and Flash, although dense, is primarily used for initialization and remains inactive during inference. These limitations highlight the need for hybrid memory organizations tailored to LLM inference. We propose Outlier-aware Quantization with Memory Co-design (QMC), a retraining-free quantization with a novel heterogeneous memory architecture. QMC identifies inlier and outlier weights in SLMs, storing inlier weights in compact multi-level Resistive-RAM (ReRAM) while preserving critical outliers in high-precision on-chip Magnetoresistive-RAM (MRAM), mitigating noise-induced degradation. On language modeling and reasoning benchmarks, QMC outperforms and matches state-of-the-art quantization methods using advanced algorithms and hybrid data formats, while achieving greater compression under both algorithm-only evaluation and realistic deployment settings. Specifically, compared against SoTA quantization methods on the latest edge AI platform, QMC reduces memory usage by 6.3x-7.3x, external data transfers by 7.6x, energy by 11.7x, and latency by 12.5x when compared to FP16, establishing QMC as a scalable, deployment-ready co-design for efficient on-device inference.

Yanru Chen, Runyang Tian, Zheyu Li et al.

While graph-based dynamic programming (DP) is a cornerstone of genomics and network analytics, its efficiency is hampered by fundamentally conflicting computational patterns. Matrix-centric DP drives regular, compute-bound network analytics, while topology-centric DP handles irregular, memory-bound genomic traversals. These two categories of DP have substantially different computation patterns and dataflows, which makes it difficult for a single homogeneous processing-in-memory (PIM) architecture to efficiently support both. This work presents GEN-Graph, a novel heterogeneous PIM chiplet that integrates two types of specialized compute tiles within a 2.5D package: Matrix-tile, a processing-using-memory (PUM) tile optimized for matrix-centric workloads, such as all-pairs shortest path (APSP); and traversal-tile, a processing-near-memory (PNM) tile optimized for traversal-centric DP workloads, such as DNA sequence alignment. Our hardware-software co-design employs recursive partitioning and reconfigurable windowed bit-parallel logic to ensure exact computation. Results show the matrix tile achieves 42.8x speedup and 392x energy efficiency over the NVIDIA H100 GPU for APSP. For sequence-to-graph alignment, the traversal tile sustains 2.56 million reads/s (short-reads) and 39.3 thousand reads/s (long-reads), outperforming state-of-the-art accelerators by up to 2.56x in throughput. GEN-Graph provides the first scalable, exact solution for general DP dataflows by matching hardware specialization to algorithmic structure.

Sean Fuhrman, Onat Gungor, Tajana Rosing

Intrusion detection systems (IDS) play a crucial role in IoT and network security by monitoring system data and alerting to suspicious activities. Machine learning (ML) has emerged as a promising solution for IDS, offering highly accurate intrusion detection. However, ML-IDS solutions often overlook two critical aspects needed to build reliable systems: continually changing data streams and a lack of attack labels. Streaming network traffic and associated cyber attacks are continually changing, which can degrade the performance of deployed ML models. Labeling attack data, such as zero-day attacks, in real-world intrusion scenarios may not be feasible, making the use of ML solutions that do not rely on attack labels necessary. To address both these challenges, we propose CND-IDS, a continual novelty detection IDS framework which consists of (i) a learning-based feature extractor that continuously updates new feature representations of the system data, and (ii) a novelty detector that identifies new cyber attacks by leveraging principal component analysis (PCA) reconstruction. Our results on realistic intrusion datasets show that CND-IDS achieves up to 6.1x F-score improvement, and up to 6.5x improved forward transfer over the SOTA unsupervised continual learning algorithm. Our code will be released upon acceptance.

Woojin Lee, Pranav Mekkoth, Ye Tian et al.

The widespread adoption of camera-equipped mobile devices and wearables has enabled convenient capture of meal images, making food recognition a key component for real time dietary monitoring. However, real-world food images present challenges due to high intra-class similarity and the frequent presence of multiple food items within a single image. While deep learning models achieve strong performance in coarse grained classification, they often struggle to capture fine-grained attributes such as cooking style. Moreover, open-ended generation in modern vision-language models can produce non-canonical labels, limiting their practical deployment. We propose FoodCHA, a multimodal agentic framework that reformulates food recognition as a hierarchical decision-making process. By progressively anchoring predictions, FoodCHA guides subcategory identification using high-level categories and guides cooking style recognition using subcategories, improving semantic consistency and attribute-level discrimination. To ensure practical deployability, FoodCHA utilizes the compact Moondream-2B vision language model, which provides strong reasoning capability while maintaining lower computational and memory overhead. Experiments on FoodNExTDB show that FoodCHA outperforms Food-Llama-3.2-11B by 13.8% and 38.2% in category and subcategory recognition precision, respectively, and achieves a striking 153.2% improvement in cooking style classification precision.

Yanru Chen, Runyang Tian, Yue Pan et al.

The proliferation of large language models (LLMs) is accelerating the integration of multimodal assistants into edge devices, where inference is executed under stringent latency and energy constraints, often exacerbated by intermittent connectivity. These challenges become particularly acute in the context of multimodal LLMs (MLLMs), as high-dimensional visual inputs are transformed into extensive token sequences, thereby inflating the key-value (KV) cache and imposing substantial data movement overheads to the LLM backbone. To address these issues, we present CHIME, a chiplet-based heterogeneous near-memory acceleration for edge MLLMs inference. CHIME leverages the complementary strengths of integrated monolithic 3D (M3D) DRAM and RRAM chiplets: DRAM supplies low-latency bandwidth for attention, while RRAM offers dense, non-volatile storage for weights. This heterogeneous hardware is orchestrated by a co-designed mapping framework that executes fused kernels near data, minimizing cross-chiplet traffic to maximize effective bandwidth. On FastVLM (0.6B/1.7B) and MobileVLM (1.7B/3B), CHIME achieves up to 54x speedup and up to 246x better energy efficiency per inference as compared to the edge GPU NVIDIA Jetson Orin NX. It sustains 116.5-266.5 token/J compared to Jetson's 0.7-1.1 token/J. Furthermore, it delivers up to 69.2x higher throughput than the state-of-the-art PIM accelerator FACIL. Compared to the M3D DRAM-only design, CHIME's heterogeneous memory further improves energy efficiency by 7% and performance by 2.4x.

Ye Tian, Jingyi Zhang, Zihao Wang et al.

Autonomous driving requires reliable reasoning over fine-grained 3D scene facts. Fine-grained question answering over multi-modal driving observations provides a natural way to evaluate this capability, yet existing perception pipelines and driving-oriented large language model (LLM) methods still suffer from unreliable scene facts, hallucinations, opaque reasoning, and heavy reliance on task-specific training. We present KLDrive, the first knowledge-graph-augmented LLM reasoning framework for fine-grained question answering in autonomous driving. KLDrive addresses this problem through designing two tightly coupled components: an energy-based scene fact construction module that consolidates multi-source evidence into a reliable scene knowledge graph, and an LLM agent that performs fact-grounded reasoning over a constrained action space under explicit structural constraints. By combining structured prompting with few-shot in-context exemplars, the framework adapts to diverse reasoning tasks without heavy task-specific fine-tuning. Experiments on two large-scale autonomous-driving QA benchmarks show that KLDrive outperforms prior state-of-the-art methods, achieving the best overall accuracy of 65.04% on NuScenes-QA and the best SPICE score of 42.45 on GVQA. On counting, the most challenging factual reasoning task, it improves over the strongest baseline by 46.01 percentage points, demonstrating substantially reduced hallucinations and the benefit of coupling reliable scene fact construction with explicit reasoning.

Ye Tian, Zihao Wang, Onat Gungor et al.

Personalized digital health support requires long-horizon, cross-dimensional reasoning over heterogeneous lifestyle signals, and recent advances in mobile sensing and large language models (LLMs) make such support increasingly feasible. However, the capabilities of current LLMs in this setting remain unclear due to the lack of systematic benchmarks. In this paper, we introduce LifeAgentBench, a large-scale QA benchmark for long-horizon, cross-dimensional, and multi-user lifestyle health reasoning, containing 22,573 questions spanning from basic retrieval to complex reasoning. We release an extensible benchmark construction pipeline and a standardized evaluation protocol to enable reliable and scalable assessment of LLM-based health assistants. We then systematically evaluate 11 leading LLMs on LifeAgentBench and identify key bottlenecks in long-horizon aggregation and cross-dimensional reasoning. Motivated by these findings, we propose LifeAgent as a strong baseline agent for health assistant that integrates multi-step evidence retrieval with deterministic aggregation, achieving significant improvements compared with two widely used baselines. Case studies further demonstrate its potential in realistic daily-life scenarios. The benchmark is publicly available at https://anonymous.4open.science/r/LifeAgentBench-CE7B.

Tianqi Zhang, Flavio Ponzina, Tajana Rosing

Approximate Nearest-Neighbor Search (ANNS) is a key technique in retrieval-augmented generation (RAG), enabling rapid identification of the most relevant high-dimensional embeddings from massive vector databases. Modern ANNS engines accelerate this process using prebuilt indexes and store compressed vector-quantized representations in fast memory. However, they still rely on a costly second-pass refinement stage that reads full-precision vectors from slower storage like SSDs. For modern text and multimodal embeddings, these reads now dominate the latency of the entire query. We propose FaTRQ, a far-memory-aware refinement system using tiered memory that eliminates the need to fetch full vectors from storage. It introduces a progressive distance estimator that refines coarse scores using compact residuals streamed from far memory. Refinement stops early once a candidate is provably outside the top-k. To support this, we propose tiered residual quantization, which encodes residuals as ternary values stored efficiently in far memory. A custom accelerator is deployed in a CXL Type-2 device to perform low-latency refinement locally. Together, FaTRQ improves the storage efficiency by 2.4$\times$ and improves the throughput by up to 9$ \times$ than SOTA GPU ANNS system.

Elvin Li, Zhengli Shang, Onat Gungor et al.

The proliferation of IoT devices has significantly increased network vulnerabilities, creating an urgent need for effective Intrusion Detection Systems (IDS). Machine Learning-based IDS (ML-IDS) offer advanced detection capabilities but rely on labeled attack data, which limits their ability to identify unknown threats. Self-Supervised Learning (SSL) presents a promising solution by using only normal data to detect patterns and anomalies. This paper introduces SAFE, a novel framework that transforms tabular network intrusion data into an image-like format, enabling Masked Autoencoders (MAEs) to learn robust representations of network behavior. The features extracted by the MAEs are then incorporated into a lightweight novelty detector, enhancing the effectiveness of anomaly detection. Experimental results demonstrate that SAFE outperforms the state-of-the-art anomaly detection method, Scale Learning-based Deep Anomaly Detection method (SLAD), by up to 26.2% and surpasses the state-of-the-art SSL-based network intrusion detection approach, Anomal-E, by up to 23.5% in F1-score.

Flavio Ponzina, Tajana Rosing

Hyperdimensional computing (HDC) is emerging as a promising AI approach that can effectively target TinyML applications thanks to its lightweight computing and memory requirements. Previous works on HDC showed that limiting the standard 10k dimensions of the hyperdimensional space to much lower values is possible, reducing even more HDC resource requirements. Similarly, other studies demonstrated that binary values can be used as elements of the generated hypervectors, leading to significant efficiency gains at the cost of some degree of accuracy degradation. Nevertheless, current optimization attempts do not concurrently co-optimize HDC hyper-parameters, and accuracy degradation is not directly controlled, resulting in sub-optimal HDC models providing several applications with unacceptable output qualities. In this work, we propose MicroHD, a novel accuracy-driven HDC optimization approach that iteratively tunes HDC hyper-parameters, reducing memory and computing requirements while ensuring user-defined accuracy levels. The proposed method can be applied to HDC implementations using different encoding functions, demonstrates good scalability for larger HDC workloads, and achieves compression and efficiency gains up to 200x when compared to baseline implementations for accuracy degradations lower than 1%.

Xiaofan Yu, Lanxiang Hu, Benjamin Reichman et al.

Natural language interaction with sensing systems is crucial for addressing users' personal concerns and providing health-related insights into their daily lives. When a user asks a question, the system automatically analyzes the full history of sensor data, extracts relevant information, and generates an appropriate response. However, existing systems are limited to short-duration (e.g., one minute) or low-frequency (e.g., daily step count) sensor data. In addition, they struggle with quantitative questions that require precise numerical answers. In this work, we introduce SensorChat, the first end-to-end QA system designed for daily life monitoring using long-duration, high-frequency time series data. Given raw sensor signals spanning multiple days and a user-defined natural language question, SensorChat generates semantically meaningful responses that directly address user concerns. SensorChat effectively handles both quantitative questions that require numerical precision and qualitative questions that require high-level reasoning to infer subjective insights. To achieve this, SensorChat uses an innovative three-stage pipeline including question decomposition, sensor data query, and answer assembly. The first and third stages leverage Large Language Models (LLMs) to interpret human queries and generate responses. The intermediate querying stage extracts relevant information from the complete sensor data history. Real-world implementations demonstrate SensorChat's capability for real-time interactions on a cloud server while also being able to run entirely on edge platforms after quantization. Comprehensive QA evaluations show that SensorChat achieves 93% higher answer accuracy than the best performing state-of-the-art systems on quantitative questions. Furthermore, a user study with eight volunteers highlights SensorChat's effectiveness in answering qualitative questions.

Ye Tian, Xiaoyuan Ren, Zihao Wang et al.

Rich and context-aware activity logs facilitate user behavior analysis and health monitoring, making them a key research focus in ubiquitous computing. The remarkable semantic understanding and generation capabilities of Large Language Models (LLMs) have recently created new opportunities for activity log generation. However, existing methods continue to exhibit notable limitations in terms of accuracy, efficiency, and semantic richness. To address these challenges, we propose DailyLLM. To the best of our knowledge, this is the first log generation and summarization system that comprehensively integrates contextual activity information across four dimensions: location, motion, environment, and physiology, using only sensors commonly available on smartphones and smartwatches. To achieve this, DailyLLM introduces a lightweight LLM-based framework that integrates structured prompting with efficient feature extraction to enable high-level activity understanding. Extensive experiments demonstrate that DailyLLM outperforms state-of-the-art (SOTA) log generation methods and can be efficiently deployed on personal computers and Raspberry Pi. Utilizing only a 1.5B-parameter LLM model, DailyLLM achieves a 17% improvement in log generation BERTScore precision compared to the 70B-parameter SOTA baseline, while delivering nearly 10x faster inference speed.

Nilesh Prasad Pandey, Shriniwas Kulkarni, David Wang et al.

Hyperdimensional Computing (HDC) is emerging as a promising approach for edge AI, offering a balance between accuracy and efficiency. However, current HDC-based applications often rely on high-precision models and/or encoding matrices to achieve competitive performance, which imposes significant computational and memory demands, especially for ultra-low power devices. While recent efforts use techniques like precision reduction and pruning to increase the efficiency, most require retraining to maintain performance, making them expensive and impractical. To address this issue, we propose a novel Post Training Compression algorithm, Decomposition-Pruning-Quantization (DPQ-HD), which aims at compressing the end-to-end HDC system, achieving near floating point performance without the need of retraining. DPQ-HD reduces computational and memory overhead by uniquely combining the above three compression techniques and efficiently adapts to hardware constraints. Additionally, we introduce an energy-efficient inference approach that progressively evaluates similarity scores such as cosine similarity and performs early exit to reduce the computation, accelerating prediction inference while maintaining accuracy. We demonstrate that DPQ-HD achieves up to 20-100x reduction in memory for image and graph classification tasks with only a 1-2% drop in accuracy compared to uncompressed workloads. Lastly, we show that DPQ-HD outperforms the existing post-training compression methods and performs better or at par with retraining-based state-of-the-art techniques, requiring significantly less overall optimization time (up to 100x) and faster inference (up to 56x) on a microcontroller

Harsh Vardhan, Xiaofan Yu, Tajana Rosing et al.

Federated learning (FL) is a distributed machine learning paradigm where multiple clients conduct local training based on their private data, then the updated models are sent to a central server for global aggregation. The practical convergence of FL is challenged by multiple factors, with the primary hurdle being the heterogeneity among clients. This heterogeneity manifests as data heterogeneity concerning local data distribution and latency heterogeneity during model transmission to the server. While prior research has introduced various efficient client selection methods to alleviate the negative impacts of either of these heterogeneities individually, efficient methods to handle real-world settings where both these heterogeneities exist simultaneously do not exist. In this paper, we propose two novel theoretically optimal client selection schemes that can handle both these heterogeneities. Our methods involve solving simple optimization problems every round obtained by minimizing the theoretical runtime to convergence. Empirical evaluations on 9 datasets with non-iid data distributions, 2 practical delay distributions, and non-convex neural network models demonstrate that our algorithms are at least competitive to and at most 20 times better than best existing baselines.

Cagla Ipek Kocal, Onat Gungor, Aaron Tartz et al.

Minimizing computational overhead in time-series classification, particularly in deep learning models, presents a significant challenge. This challenge is further compounded by adversarial attacks, emphasizing the need for resilient methods that ensure robust performance and efficient model selection. We introduce ReLATE, a framework that identifies robust learners based on dataset similarity, reduces computational overhead, and enhances resilience. ReLATE maintains multiple deep learning models in well-known adversarial attack scenarios, capturing model performance. ReLATE identifies the most analogous dataset to a given target using a similarity metric, then applies the optimal model from the most similar dataset. ReLATE reduces computational overhead by an average of 81.2%, enhancing adversarial resilience and streamlining robust model selection, all without sacrificing performance, within 4.2% of Oracle.

Onat Gungor, Amanda Sofie Rios, Nilesh Ahuja et al.

Detecting out-of-distribution (OOD) data is a fundamental challenge in the deployment of machine learning models. From a security standpoint, this is particularly important because OOD test data can result in misleadingly confident yet erroneous predictions, which undermine the reliability of the deployed model. Although numerous models for OOD detection have been developed in computer vision and language, their adaptability to the time-series data domain remains limited and under-explored. Yet, time-series data is ubiquitous across manufacturing and security applications for which OOD is essential. This paper seeks to address this research gap by conducting a comprehensive analysis of modality-agnostic OOD detection algorithms. We evaluate over several multivariate time-series datasets, deep learning architectures, time-series specific data augmentations, and loss functions. Our results demonstrate that: 1) the majority of state-of-the-art OOD methods exhibit limited performance on time-series data, and 2) OOD methods based on deep feature modeling may offer greater advantages for time-series OOD detection, highlighting a promising direction for future time-series OOD detection algorithm development.

Esha Singh, Dongxia Wu, Chien-Yi Yang et al.

Multi-objective combinatorial optimization seeks Pareto-optimal solutions over exponentially large discrete spaces, yet existing methods sacrifice generality, scalability, or theoretical guarantees. We reformulate it as an online learning problem over a decomposed decision space, solving position-wise bandit subproblems via adaptive expert-guided sequential construction. This formulation admits regret bounds of $O(d\sqrt{T \log T})$ depending on subproblem dimensionality \(d\) rather than combinatorial space size. On standard benchmarks, our method achieves 80--98\% of specialized solvers performance while achieving two to three orders of magnitude improvement in sample and computational efficiency over Bayesian optimization methods. On real-world hardware-software co-design for AI accelerators with expensive simulations, we outperform competing methods under fixed evaluation budgets. The advantage grows with problem scale and objective count, establishing bandit optimization over decomposed decision spaces as a principled alternative to surrogate modeling or offline training for multi-objective optimization.

Yue Pan, Zihan Xia, Po-Kai Hsu et al.

As Large Language Models (LLMs) continue to evolve, Mixture of Experts (MoE) architecture has emerged as a prevailing design for achieving state-of-the-art performance across a wide range of tasks. MoE models use sparse gating to activate only a handful of expert sub-networks per input, achieving billion-parameter capacity with inference costs akin to much smaller models. However, such models often pose challenges for hardware deployment due to the massive data volume introduced by the MoE layers. To address the challenges of serving MoE models, we propose Stratum, a system-hardware co-design approach that combines the novel memory technology Monolithic 3D-Stackable DRAM (Mono3D DRAM), near-memory processing (NMP), and GPU acceleration. The logic and Mono3D DRAM dies are connected through hybrid bonding, whereas the Mono3D DRAM stack and GPU are interconnected via silicon interposer. Mono3D DRAM offers higher internal bandwidth than HBM thanks to the dense vertical interconnect pitch enabled by its monolithic structure, which supports implementations of higher-performance near-memory processing. Furthermore, we tackle the latency differences introduced by aggressive vertical scaling of Mono3D DRAM along the z-dimension by constructing internal memory tiers and assigning data across layers based on access likelihood, guided by topic-based expert usage prediction to boost NMP throughput. The Stratum system achieves up to 8.29x improvement in decoding throughput and 7.66x better energy efficiency across various benchmarks compared to GPU baselines.

Daniel Zhao, Abhilash Shankarampeta, Lanxiang Hu et al.

We propose a novel method that leverages sparse autoencoders (SAEs) and clustering techniques to analyze the internal token representations of large language models (LLMs) and guide generations in mathematical reasoning tasks. Our approach first trains an SAE to generate sparse vector representations for training tokens, then applies k-means clustering to construct a graph where vertices represent token clusters and weighted edges capture sequential token transitions. Using this graph, we define an edge-weight based reward function to quantify adherence to established reasoning traces, thereby identifying exploitative reasoning trajectories. Additionally, we measure generation diversity from clustering to assess the extent of exploration. Our findings indicate that balancing both exploitation and exploration is crucial for achieving high accuracy in mathematical reasoning tasks. During generation, the SAE can serve as a scalable reward model to guide generations, ensuring a balanced trade-off between exploitation and exploration. This prevents extreme behaviors in either direction, ultimately fostering a higher-quality reasoning process in LLMs.

You Hak Lee, Xiaofan Yu, Quanling Zhao et al.

Unsupervised federated learning (UFL) has gained attention as a privacy-preserving, decentralized machine learning approach that eliminates the need for labor-intensive data labeling. However, UFL faces several challenges in practical applications: (1) non-independent and identically distributed (non-iid) data distribution across devices, (2) expensive computational and communication costs at the edge, and (3) vulnerability to communication noise. Previous UFL approaches have relied on deep neural networks (NN), which introduce substantial overhead in both computation and communication. In this paper, we propose FedUHD, the first UFL framework based on Hyperdimensional Computing (HDC). HDC is a brain-inspired computing scheme with lightweight training and inference operations, much smaller model size, and robustness to communication noise. FedUHD introduces two novel HDC-based designs to improve UFL performance. On the client side, a kNN-based cluster hypervector removal method addresses non-iid data samples by eliminating detrimental outliers. On the server side, a weighted HDC aggregation technique balances the non-iid data distribution across clients. Our experiments demonstrate that FedUHD achieves up to 173.6x and 612.7x better speedup and energy efficiency, respectively, in training, up to 271x lower communication cost, and 15.50% higher accuracy on average across diverse settings, along with superior robustness to various types of noise compared to state-of-the-art NN-based UFL approaches.

Chang Eun Song, Weihong Xu, Keming Fan et al.

Clo-HDnn is an on-device learning (ODL) accelerator designed for emerging continual learning (CL) tasks. Clo-HDnn integrates hyperdimensional computing (HDC) along with low-cost Kronecker HD Encoder and weight clustering feature extraction (WCFE) to optimize accuracy and efficiency. Clo-HDnn adopts gradient-free CL to efficiently update and store the learned knowledge in the form of class hypervectors. Its dual-mode operation enables bypassing costly feature extraction for simpler datasets, while progressive search reduces complexity by up to 61% by encoding and comparing only partial query hypervectors. Achieving 4.66 TFLOPS/W (FE) and 3.78 TOPS/W (classifier), Clo-HDnn delivers 7.77x and 4.85x higher energy efficiency compared to SOTA ODL accelerators.

Le Zhang, Quanling Zhao, Run Wang et al.

Learning-based environmental sound recognition has emerged as a crucial method for ultra-low-power environmental monitoring in biological research and city-scale sensing systems. These systems usually operate under limited resources and are often powered by harvested energy in remote areas. Recent efforts in on-device sound recognition suffer from low accuracy due to resource constraints, whereas cloud offloading strategies are hindered by high communication costs. In this work, we introduce ORCA, a novel resource-efficient cloud-assisted environmental sound recognition system on batteryless devices operating over the Low-Power Wide-Area Networks (LPWANs), targeting wide-area audio sensing applications. We propose a cloud assistance strategy that remedies the low accuracy of on-device inference while minimizing the communication costs for cloud offloading. By leveraging a self-attention-based cloud sub-spectral feature selection method to facilitate efficient on-device inference, ORCA resolves three key challenges for resource-constrained cloud offloading over LPWANs: 1) high communication costs and low data rates, 2) dynamic wireless channel conditions, and 3) unreliable offloading. We implement ORCA on an energy-harvesting batteryless microcontroller and evaluate it in a real world urban sound testbed. Our results show that ORCA outperforms state-of-the-art methods by up to $80 \times$ in energy savings and $220 \times$ in latency reduction while maintaining comparable accuracy.

Lanxiang Hu, Tajana Rosing, Hao Zhang

Specializing large language models (LLMs) for local deployment in domain-specific use cases is necessary for strong performance while meeting latency and privacy constraints. However, conventional task-specific adaptation approaches do not show simultaneous memory saving and inference speedup at deployment time. Practical compression techniques like quantization and pruning require dedicated hardware or kernel support to achieve measured inference speedup. We develop TrimLLM based on the layer-wise specialization phenomenon we empirically observed and verified on contemporary LLMs. TrimLLM reduces the depth of LLMs via progressive layer dropping. We show it retains LLMs' capacity in specific domains and achieves inference speedup irrespective of hardware and deep learning frameworks. We evaluated TrimLLM on LLMs of various sizes for inference; models adapted on medical, legal, and financial datasets all demonstrate $2.1-5.7\times$ inference speedup on consumer GPUs and up to $3.1\times$ speedup on A100 when compared to state-of-the-art model compression algorithms, with no loss in accuracy at 50$\sim$60\% model compression ratio.

Kazim Ergun, Rishikanth Chandrasekaran, Tajana Rosing

Federated learning (FL) enables a loose set of participating clients to collaboratively learn a global model via coordination by a central server and with no need for data sharing. Existing FL approaches that rely on complex algorithms with massive models, such as deep neural networks (DNNs), suffer from computation and communication bottlenecks. In this paper, we first propose FedHDC, a federated learning framework based on hyperdimensional computing (HDC). FedHDC allows for fast and light-weight local training on clients, provides robust learning, and has smaller model communication overhead compared to learning with DNNs. However, current HDC algorithms get poor accuracy when classifying larger & more complex images, such as CIFAR10. To address this issue, we design FHDnn, which complements FedHDC with a self-supervised contrastive learning feature extractor. We avoid the transmission of the DNN and instead train only the HDC learner in a federated manner, which accelerates learning, reduces transmission cost, and utilizes the robustness of HDC to tackle network errors. We present a formal analysis of the algorithm and derive its convergence rate both theoretically, and show experimentally that FHDnn converges 3$\times$ faster vs. DNNs. The strategies we propose to improve the communication efficiency enable our design to reduce communication costs by 66$\times$ vs. DNNs, local client compute and energy consumption by ~1.5 - 6$\times$, while being highly robust to network errors. Finally, our proposed strategies for improving the communication efficiency have up to 32$\times$ lower communication costs with good accuracy.

Gopi Krishna Jha, Anthony Thomas, Nilesh Jain et al.

Deep learning-based recommendation systems (e.g., DLRMs) are widely used AI models to provide high-quality personalized recommendations. Training data used for modern recommendation systems commonly includes categorical features taking on tens-of-millions of possible distinct values. These categorical tokens are typically assigned learned vector representations, that are stored in large embedding tables, on the order of 100s of GB. Storing and accessing these tables represent a substantial burden in commercial deployments. Our work proposes MEM-REC, a novel alternative representation approach for embedding tables. MEM-REC leverages bloom filters and hashing methods to encode categorical features using two cache-friendly embedding tables. The first table (token embedding) contains raw embeddings (i.e. learned vector representation), and the second table (weight embedding), which is much smaller, contains weights to scale these raw embeddings to provide better discriminative capability to each data point. We provide a detailed architecture, design and analysis of MEM-REC addressing trade-offs in accuracy and computation requirements, in comparison with state-of-the-art techniques. We show that MEM-REC can not only maintain the recommendation quality and significantly reduce the memory footprint for commercial scale recommendation models but can also improve the embedding latency. In particular, based on our results, MEM-REC compresses the MLPerf CriteoTB benchmark DLRM model size by 2900x and performs up to 3.4x faster embeddings while achieving the same AUC as that of the full uncompressed model.

Anthony Thomas, Sanjoy Dasgupta, Tajana Rosing

Hyperdimensional (HD) computing is a set of neurally inspired methods for obtaining high-dimensional, low-precision, distributed representations of data. These representations can be combined with simple, neurally plausible algorithms to effect a variety of information processing tasks. HD computing has recently garnered significant interest from the computer hardware community as an energy-efficient, low-latency, and noise-robust tool for solving learning problems. In this review, we present a unified treatment of the theoretical foundations of HD computing with a focus on the suitability of representations for learning.

Behnam Khaleghi, Sahand Salamat, Anthony Thomas et al.

Hyperdimensional computing (HD) is an emerging paradigm for machine learning based on the evidence that the brain computes on high-dimensional, distributed, representations of data. The main operation of HD is encoding, which transfers the input data to hyperspace by mapping each input feature to a hypervector, accompanied by so-called bundling procedure that simply adds up the hypervectors to realize encoding hypervector. Although the operations of HD are highly parallelizable, the massive number of operations hampers the efficiency of HD in embedded domain. In this paper, we propose SHEARer, an algorithm-hardware co-optimization to improve the performance and energy consumption of HD computing. We gain insight from a prudent scheme of approximating the hypervectors that, thanks to inherent error resiliency of HD, has minimal impact on accuracy while provides high prospect for hardware optimization. In contrast to previous works that generate the encoding hypervectors in full precision and then ex-post quantizing, we compute the encoding hypervectors in an approximate manner that saves a significant amount of resources yet affords high accuracy. We also propose a novel FPGA implementation that achieves striking performance through massive parallelism with low power consumption. Moreover, we develop a software framework that enables training HD models by emulating the proposed approximate encodings. The FPGA implementation of SHEARer achieves an average throughput boost of 104,904x (15.7x) and energy savings of up to 56,044x (301x) compared to state-of-the-art encoding methods implemented on Raspberry Pi 3 (GeForce GTX 1080 Ti) using practical machine learning datasets.

Behnam Khaleghi, Mohsen Imani, Tajana Rosing

The privacy of data is a major challenge in machine learning as a trained model may expose sensitive information of the enclosed dataset. Besides, the limited computation capability and capacity of edge devices have made cloud-hosted inference inevitable. Sending private information to remote servers makes the privacy of inference also vulnerable because of susceptible communication channels or even untrustworthy hosts. In this paper, we target privacy-preserving training and inference of brain-inspired Hyperdimensional (HD) computing, a new learning algorithm that is gaining traction due to its light-weight computation and robustness particularly appealing for edge devices with tight constraints. Indeed, despite its promising attributes, HD computing has virtually no privacy due to its reversible computation. We present an accuracy-privacy trade-off method through meticulous quantization and pruning of hypervectors, the building blocks of HD, to realize a differentially private model as well as to obfuscate the information sent for cloud-hosted inference. Finally, we show how the proposed techniques can be also leveraged for efficient hardware implementation.

Yunhui Guo, Noel C. Codella, Leonid Karlinsky et al.

Recent progress on few-shot learning largely relies on annotated data for meta-learning: base classes sampled from the same domain as the novel classes. However, in many applications, collecting data for meta-learning is infeasible or impossible. This leads to the cross-domain few-shot learning problem, where there is a large shift between base and novel class domains. While investigations of the cross-domain few-shot scenario exist, these works are limited to natural images that still contain a high degree of visual similarity. No work yet exists that examines few-shot learning across different imaging methods seen in real world scenarios, such as aerial and medical imaging. In this paper, we propose the Broader Study of Cross-Domain Few-Shot Learning (BSCD-FSL) benchmark, consisting of image data from a diverse assortment of image acquisition methods. This includes natural images, such as crop disease images, but additionally those that present with an increasing dissimilarity to natural images, such as satellite images, dermatology images, and radiology images. Extensive experiments on the proposed benchmark are performed to evaluate state-of-art meta-learning approaches, transfer learning approaches, and newer methods for cross-domain few-shot learning. The results demonstrate that state-of-art meta-learning methods are surprisingly outperformed by earlier meta-learning approaches, and all meta-learning methods underperform in relation to simple fine-tuning by 12.8% average accuracy. Performance gains previously observed with methods specialized for cross-domain few-shot learning vanish in this more challenging benchmark. Finally, accuracy of all methods tend to correlate with dataset similarity to natural images, verifying the value of the benchmark to better represent the diversity of data seen in practice and guiding future research.