Search for: All records

Contrastive learning (CL) has been widely investigated with various learning mech- anisms and achieves strong capability in learning representations of data in a self-supervised manner using unlabeled data. A common fashion of contrastive learning on this line is employing large-sized encoders to achieve comparable performance as the supervised learning counterpart. Despite the success of the labelless training, current contrastive learning algorithms failed to achieve good performance with lightweight (compact) models, e.g., MobileNet, while the re- quirements of the heavy encoders impede the energy-efficient computation, espe- cially for resource-constrained AI applications. Motivated by this, we propose a new self-supervised CL scheme, named SACL-XD, consisting of two technical components, Slimmed Asymmetrical Contrastive Learning (SACL) and Cross- Distillation (XD), which collectively enable efficient CL with compact models. While relevant prior works employed a strong pre-trained model as the teacher of unsupervised knowledge distillation to a lightweight encoder, our proposed method trains CL models from scratch and outperforms them even without such an expensive requirement. Compared to the SoTA lightweight CL training (dis- tillation) algorithms, SACL-XD achieves 1.79% ImageNet-1K accuracy improve- ment on MobileNet-V3 with 64⇥ training FLOPs reduction. Code is available at https://github.com/mengjian0502/SACL-XD. 

In genomic analysis, the major computation bottle- neck is the memory- and compute-intensive DNA short reads alignment due to memory-wall challenge. This work presents the first Resistive RAM (RRAM) based Compute-in-Memory (CIM) macro design for accelerating state-of-the-art BWT based genome sequencing alignment. Our design could support all the core instructions, i.e., XNOR based match, count, and addition, required by alignment algorithm. The proposed CIM macro implemented in integration of HfO2 RRAM and 65nm CMOS demonstrates the best energy efficiency to date with 2.07 TOPS/W and 2.12G suffixes/J at 1.0V. 

Channel decoders are key computing modules in wired/wireless communication systems. Recently neural network (NN)-based decoders have shown their promising error-correcting performance because of their end-to-end learning capability. However, compared with the traditional approaches, the emerging neural belief propagation (NBP) solution suffers higher storage and computational complexity, limiting its hardware performance. To address this challenge and develop a channel decoder that can achieve high decoding performance and hardware performance simultaneously, in this paper we take a first step towards exploring SRAM-based in-memory computing for efficient NBP channel decoding. We first analyze the unique sparsity pattern in the NBP processing, and then propose an efficient and fully Digital Sparse In-Memory Matrix vector Multiplier (DSPIMM) computing platform. Extensive experiments demonstrate that our proposed DSPIMM achieves significantly higher energy efficiency and throughput than the state-of-the-art counterparts. 

Deep Neural Network (DNN) acceleration with digital Processing-in-Memory (PIM) platforms at the edge is an actively-explored domain with great potential to not only address memory-wall bottlenecks but to offer orders of performance improvement in comparison to the von-Neumann architecture. On the other side, FPGA-based edge computing has been followed as a potential solution to accelerate compute-intensive workloads. In this work, adopting low-bit-width neural networks, we perform a solid and comparative inference performance analysis of a recent processing-in-SRAM tape-out with a low-resource FPGA board and a high-performance GPU to provide a guideline for the research community. We explore and highlight the key architectural constraints of these edge candidates that impact their overall performance. Our experimental data demonstrate that the processing-in-SRAM can obtain up to ~160x speed-up and up to 228x higher efficiency (img/s/W) compared to the under-test FPGA on the CIFAR-10 dataset. 

By learning a sequence of tasks continually, an agent in continual learning (CL) can improve the learning performance of both a new task and `old' tasks by leveraging the forward knowledge transfer and the backward knowledge transfer, respectively. However, most existing CL methods focus on addressing catastrophic forgetting in neural networks by minimizing the modification of the learnt model for old tasks. This inevitably limits the backward knowledge transfer from the new task to the old tasks, because judicious model updates could possibly improve the learning performance of the old tasks as well. To tackle this problem, we first theoretically analyze the conditions under which updating the learnt model of old tasks could be beneficial for CL and also lead to backward knowledge transfer, based on the gradient projection onto the input subspaces of old tasks. Building on the theoretical analysis, we next develop a ContinUal learning method with Backward knowlEdge tRansfer (CUBER), for a fixed capacity neural network without data replay. In particular, CUBER first characterizes the task correlation to identify the positively correlated old tasks in a layer-wise manner, and then selectively modifies the learnt model of the old tasks when learning the new task. Experimental studies show that CUBER can even achieve positive backward knowledge transfer on several existing CL benchmarks for the first time without data replay, where the related baselines still suffer from catastrophic forgetting (negative backward knowledge transfer). The superior performance of CUBER on the backward knowledge transfer also leads to higher accuracy accordingly. 

Recently, a new trend of exploring training sparsity has emerged, which remove parameters during training, leading to both training and inference efficiency improvement. This line of works primarily aims to obtain a single sparse model under a pre-defined large sparsity ratio. It leads to a static/fixed sparse inference model that is not capable of adjusting or re-configuring its computation complexity (i.e., inference structure, latency) after training for real-world varying and dynamic hardware resource availability. To enable such run-time or post-training network morphing, the concept of `dynamic inference' or `training-once-for-all' has been proposed to train a single network consisting of multiple sub-nets once, but each sub-net could perform the same inference function with different computing complexity. However, the traditional dynamic inference training method requires a joint training scheme with multi-objective optimization, which suffers from very large training overhead. In this work, for the first time, we propose a novel alternating sparse training (AST) scheme to train multiple sparse sub-nets for dynamic inference without extra training cost compared to the case of training a single sparse model from scratch. Furthermore, to mitigate the interference of weight update among sub-nets, we propose gradient correction within the inner-group iterations to reduce their weight update interference. We validate the proposed AST on multiple datasets against state-of-the-art sparse training method, which shows that AST achieves similar or better accuracy, but only needs to train once to get multiple sparse sub-nets with different sparsity ratios. More importantly, compared with the traditional joint training based dynamic inference training methodology, the large training overhead is completely eliminated without affecting the accuracy of each sub-net. 

Recently, ReRAM crossbar-based deep neural network (DNN) accelerator has been widely investigated. However, most prior works focus on single-task inference due to the high energy consumption of weight reprogramming and ReRAM cells’ low endurance issue. Adapting the ReRAM crossbar-based DNN accelerator for multiple tasks has not been fully explored. In this study, we propose XMA 2 , a novel crossbar-aware learning method with a 2-tier masking technique to efficiently adapt a DNN backbone model deployed in the ReRAM crossbar for new task learning. During the XMA 2 -based multi-task adaption (MTA), the tier-1 ReRAM crossbar-based processing-element- (PE-) wise mask is first learned to identify the most critical PEs to be reprogrammed for essential new features of the new task. Subsequently, the tier-2 crossbar column-wise mask is applied within the rest of the weight-frozen PEs to learn a hardware-friendly and column-wise scaling factor for new task learning without modifying the weight values. With such crossbar-aware design innovations, we could implement the required masking operation in an existing crossbar-based convolution engine with minimal hardware/memory overhead to adapt to a new task. The extensive experimental results show that compared with other state-of-the-art multiple-task adaption methods, XMA 2 achieves the highest accuracy on all popular multi-task learning datasets. 

Nowadays, one practical limitation of deep neural network (DNN) is its high degree of specialization to a single task or domain (e.g., one visual domain). It motivates researchers to develop algorithms that can adapt DNN model to multiple domains sequentially, while still performing well on the past domains, which is known as multi-domain learning. Almost all conventional methods only focus on improving accuracy with minimal parameter update, while ignoring high computing and memory cost during training, which makes it difficult to deploy multi-domain learning into more and more widely used resource-limited edge devices, like mobile phone, IoT, embedded system, etc. During our study in multi-domain training process, we observe that large memory used for activation storage is the bottleneck that largely limits the training time and cost on edge devices. To reduce training memory usage, while keeping the domain adaption accuracy performance, we propose Dynamic Additive Attention Adaption (DA3), a novel memory-efficient on-device multi-domain learning method. DA3 learns a novel additive attention adaptor module, while freezing the weights of the pre-trained backbone model for each domain. Differentiating from prior works, such module not only mitigates activation memory buffering for reducing memory usage during training, but also serves as a dynamic gating mechanism to reduce the computation cost for fast inference. We validate DA3 on multiple datasets against state-of-the-art methods, which shows great improvement in both accuracy and training time. Moreover, we deployed DA3 into the popular NIVDIA Jetson Nano edge GPU, where the measured experimental results show our proposed \mldam reduces the on-device training memory consumption by 19x-37x, and training time by 2x, in comparison to the baseline methods (e.g., standard fine-tuning, Parallel and Series Res. adaptor, and Piggyback). 

We present a generic and programmable Processing-in-SRAM (PSRAM) accelerator chip design based on an 8T-SRAM array to accommodate a complete set of Boolean logic operations (e.g., NOR/NAND/XOR, both 2- and 3-input), majority, and full adder, for the first time, all in a single cycle. PSRAM provides the programmability required for in-memory computing platforms that could be used for various applications such as parallel vector operation, neural networks, and data encryption. The prototype design is implemented in a SRAM macro with size of 16 kb, demonstrating one of the fastest programmable in-memory computing system to date operating at 1.23 GHz. The 65nm prototype chip achieves system-level peak throughput of 1.2 TOPS, and energy-efficiency of 34.98 TOPS/W at 1.2V.