BZPEERPapers, people, evidence and research paths
Menu

Discover

SearchFind papers, people, topics, institutions and opportunities from one place.GraphSee how papers, authors, topics and evidence connect.

Research tools

HomeContinue from the papers, people and topics that matter now.CollectionsTurn saved papers into reading lists, evidence maps and shareable context.ResearchReconstruct the path that led you from search to paper to question.CollaborationMove from a paper, topic or expert request into a focused thread.PublishCreate full-text research articles with metadata, formulas and versions.

Network

ExpertsFind specialists whose work and reviews explain why they can help.CommunitiesJoin research groups organized around papers, topics and shared evidence.

Opportunities

ListingsBrowse roles, calls and collaborations connected to real research context.

Account

Log inReturn to your saved papers, graph views and active threads.Create accountStart saving papers, building a research trail and joining teams.
Log inCreate account

Source author record

Yunji Chen

Yunji Chen appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record
Machine LearningArtificial IntelligenceComputer VisionHardware ArchitectureComputation and LanguageComputational Complexityeess.IVMultiagent SystemsNetworking and Internet ArchitectureOperating SystemsProgramming Languages

Catalog footprint

What is connected

9works
11topics
4close collaborators

Actions

Connect this record

Log in to claim
Open graphBrowse works

Research graph

See the researcher in context

Open full explorer

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

Published work

9 published item(s)

preprint2026arXiv

Discrete Diffusion for Complex and Congested Multi-Agent Path Finding with Sparse Social Attention

Multi-Agent Path Finding (MAPF) is a coordination problem that requires computing globally consistent, collision-free trajectories from individual start positions to assigned goal positions under combinatorial planning complexity. In dense environments, suboptimal initial plans induce compound conflicts that hinder feasible repair. For repair-based solvers like LNS2, initial plan quality critically affects downstream repair, yet this factor remains underexplored. We propose DiffLNS, a hybrid framework that integrates a discrete denoising diffusion probabilistic model (D3PM) with LNS2. The D3PM serves as an initializer with sparse social attention that learns a spatiotemporal prior over coordinated multi-agent action trajectories from expert demonstrations and samples multiple joint plans. Operating directly on the categorical action space, our discrete diffusion preserves the MAPF action structure and samples from a multimodal joint-plan distribution to produce diverse drafts well suited for neighborhood repair. These drafts act as warm starts for downstream repair, which completes unfinished trajectories and resolves remaining conflicts under hard MAPF constraints. Experimental results show that despite being trained only on instances with at most 96 agents, the initializer generalizes to scenarios with up to 312 agents at inference time. Across 20 complex and congested settings, DiffLNS achieves an average success rate of 95.8%, outperforming the strongest tested baseline by 9.6 percentage points and matching or exceeding all baselines in all 20 settings. To the best of our knowledge, this is the first work to leverage discrete diffusion for warm-starting an LNS-based MAPF solver.

0 citations0 reviewsNo code linkedOpen access
preprint2026arXiv

Hardwired-Neurons Language Processing Units as General-Purpose Cognitive Substrates

The rapid advancement of Large Language Models (LLMs) has established language as a core general-purpose cognitive substrate, driving the demand for specialized Language Processing Units (LPUs) tailored for LLM inference. To overcome the growing energy consumption of LLM inference systems, this paper proposes a Hardwired-Neurons Language Processing Unit (HNLPU), which physically hardwires LLM weight parameters into the computational fabric, achieving several orders of magnitude computational efficiency improvement by extreme specialization. However, a significant challenge still lies in the scale of modern LLMs. A straightforward hardwiring of gpt-oss 120 B would require fabricating photomask sets valued at over 6 billion dollars, rendering this straightforward solution economically impractical. Addressing this challenge, we propose the novel Metal-Embedding methodology. Instead of embedding weights in a 2D grid of silicon device cells, Metal-Embedding embeds weight parameters into the 3D topology of metal wires. This brings two benefits: (1) a 15x increase in density, and (2) 60 out of 70 photomask layers are homogeneous across chips, including all EUV photomasks. In total, Metal-Embedding reduced the photomask cost by 112x, bringing the Non-Recurring Engineering (NRE) cost of HNLPU into an economically viable range. Experimental results show that HNLPU achieved 249,960 tokens/s (5,555x/85x that of GPU/WSE), 36 tokens/J (1,047x/283x that of GPU/WSE), 13,232 mm2 total die area, $59.46 M-123.5 M estimated NRE at 5 nm technology. Analysis shows that HNLPU achieved 41.7-80.4x improvement in cost-effectiveness and 357x reduction in carbon footprint compared to OpenAI-scale H100 clusters, under an annual weight updating assumption.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Neural Program Synthesis with Query

Aiming to find a program satisfying the user intent given input-output examples, program synthesis has attracted increasing interest in the area of machine learning. Despite the promising performance of existing methods, most of their success comes from the privileged information of well-designed input-output examples. However, providing such input-output examples is unrealistic because it requires the users to have the ability to describe the underlying program with a few input-output examples under the training distribution. In this work, we propose a query-based framework that trains a query neural network to generate informative input-output examples automatically and interactively from a large query space. The quality of the query depends on the amount of the mutual information between the query and the corresponding program, which can guide the optimization of the query framework. To estimate the mutual information more accurately, we introduce the functional space (F-space) which models the relevance between the input-output examples and the programs in a differentiable way. We evaluate the effectiveness and generalization of the proposed query-based framework on the Karel task and the list processing task. Experimental results show that the query-based framework can generate informative input-output examples which achieve and even outperform well-designed input-output examples.

0 citations0 reviewsNo code linkedOpen access
preprint2020arXiv

Adaptive Precision Training: Quantify Back Propagation in Neural Networks with Fixed-point Numbers

Adaptive Precision Training: Quantify Back Propagation in Neural Networks with Fixed-point Numbers. Recent emerged quantization technique has been applied to inference of deep neural networks for fast and efficient execution. However, directly applying quantization in training can cause significant accuracy loss, thus remaining an open challenge.

0 citations0 reviewsNo code linkedOpen access
preprint2020arXiv

DWM: A Decomposable Winograd Method for Convolution Acceleration

Winograd's minimal filtering algorithm has been widely used in Convolutional Neural Networks (CNNs) to reduce the number of multiplications for faster processing. However, it is only effective on convolutions with kernel size as 3x3 and stride as 1, because it suffers from significantly increased FLOPs and numerical accuracy problem for kernel size larger than 3x3 and fails on convolution with stride larger than 1. In this paper, we propose a novel Decomposable Winograd Method (DWM), which breaks through the limitation of original Winograd's minimal filtering algorithm to a wide and general convolutions. DWM decomposes kernels with large size or large stride to several small kernels with stride as 1 for further applying Winograd method, so that DWM can reduce the number of multiplications while keeping the numerical accuracy. It enables the fast exploring of larger kernel size and larger stride value in CNNs for high performance and accuracy and even the potential for new CNNs. Comparing against the original Winograd, the proposed DWM is able to support all kinds of convolutions with a speedup of ~2, without affecting the numerical accuracy.

0 citations0 reviewsNo code linkedOpen access
preprint2012arXiv

DLS: Directoryless Shared Last-level Cache

Directory-based protocols have been the de facto solution for maintaining cache coherence in shared-memory parallel systems comprising multi/many cores, where each store instruction is eagerly made globally visible by invalidating the private cache (PC) backups of other cores. Consequently, the directory not only consumes large chip area, but also incurs considerable energy consumption and performance degradation, due to the large number of Invalidation/Ack messages transferred in the interconnection network and resulting network congestion. In this paper, we reveal the interesting fact that the directory is actually an unnecessary luxury for practical parallel systems. Because of widely deployed software/hardware techniques involving instruction reordering, most (if not all) parallel systems work under the weak consistency model, where a remote store instruction is allowed to be invisible to a core before the next synchronization of the core, instead of being made visible eagerly by invalidating PC backups of other cores. Based on this key observation, we propose a lightweight novel scheme called {\em DLS (DirectoryLess Shared last-level cache)}, which completely removes the directory and Invalidation/Ack messages, and efficiently maintains cache coherence using a novel {\em self-suspicion + speculative execution} mechanism. Experimental results over SPLASH-2 benchmarks show that on a 16-core processor, DLS not only completely removes the chip area cost of the directory, but also improves processor performance by 11.08%, reduces overall network traffic by 28.83%, and reduces energy consumption of the network by 15.65% on average (compared with traditional MESI protocol with full directory). Moreover, DLS does not involve any modification to programming languages and compilers, and hence is seamlessly compatible with legacy codes.

0 citations0 reviewsNo code linkedOpen access
preprint2012arXiv

Global Adaptive Routing Algorithm Without Additional Congestion Propagation Network

Adaptive routing algorithm has been employed in multichip interconnection networks in order to improve network performance. Does a algorithm use local or global network state? This is the key question in adaptive routing. In many traffic patterns, the ignorance of global network state, leading to routing selection based only on local congestion information, tends to violate global load balance. To attack the load balance issue in adapting routing, some global adaptive routing algorithms introduce a congestion propagation network to obtain global network status information, such as Regional Congestion Awareness (RCA) and Destination Based Adaptive Routing (DBAR). However, the congestion propagation network leads to additional power and area consumption which cannot be ignored. From another view, if we just increase the bandwidth between neighbor nodes with the wires used to build the congestion propagation network, the network performance could be improved as well. In this paper, we propose a global adaptive routing algorithm without employing the additional congestion propagation network. Our algorithm obtains the global network state in a novel way, and can offer significant improvement than the base-line local adaptive routing algorithm (xy-adaptive algorithm which selects routing based on local congestion information in each hop) for both medium and high injection rates. In wormhole flow control, all the routing information (flit id, source node id, destination node id, vc id and address) is contained in head flit, and data is carried in body flits. As a result, there are always many free bits in the head flit, especially when the bandwidth is 128-bits which is normal in interconnection network design. Then, we can use these free bits in the head flit to propagate global congestion information but not increase the number of flits.

0 citations0 reviewsNo code linkedOpen access
preprint2011arXiv

Efficient Deterministic Replay Using Complete Race Detection

Data races can significantly affect the executions of multi-threaded programs. Hence, one has to recur the results of data races to deterministically replay a multi-threaded program. However, data races are concealed in enormous number of memory operations in a program. Due to the difficulty of accurately identifying data races, previous multi-threaded deterministic record/replay schemes for commodity multi-processor system give up to record data races directly. Consequently, they either record all shared memory operations, which brings remarkable slowdown to the production run, or record the synchronization only, which introduces significant efforts to replay. Inspired by the advances in data race detection, we propose an efficient software-only deterministic replay scheme for commodity multi-processor systems, which is named RacX. The key insight of RacX is as follows: although it is NP-hard to accurately identify the existence of data races between a pair of memory operations, we can find out all potential data races in a multi-threaded program, in which the false positives can be reduced to a small amount with our automatic false positive reduction techniques. As a result, RacX can efficiently monitor all potential data races to deterministically replay a multi-threaded program. To evaluate RacX, we have carried out experiments over a number of well-known multi-threaded programs from SPLASH-2 benchmark suite and large-scale commercial programs. RacX can precisely recur production runs of these programs with value determinism. Averagely, RacX causes only about 1.21%, 1.89%, 2.20%, and 8.41% slowdown to the original run during recording (for 2-, 4-, 8- and 16-thread programs, respectively). The soundness, efficiency, scalability, and portability of RacX well demonstrate its superiority.

0 citations0 reviewsNo code linkedOpen access
preprint2011arXiv

The Impact of Mutation Rate on the Computation Time of Evolutionary Dynamic Optimization

Mutation has traditionally been regarded as an important operator in evolutionary algorithms. In particular, there have been many experimental studies which showed the effectiveness of adapting mutation rates for various static optimization problems. Given the perceived effectiveness of adaptive and self-adaptive mutation for static optimization problems, there have been speculations that adaptive and self-adaptive mutation can benefit dynamic optimization problems even more since adaptation and self-adaptation are capable of following a dynamic environment. However, few theoretical results are available in analyzing rigorously evolutionary algorithms for dynamic optimization problems. It is unclear when adaptive and self-adaptive mutation rates are likely to be useful for evolutionary algorithms in solving dynamic optimization problems. This paper provides the first rigorous analysis of adaptive mutation and its impact on the computation times of evolutionary algorithms in solving certain dynamic optimization problems. More specifically, for both individual-based and population-based EAs, we have shown that any time-variable mutation rate scheme will not significantly outperform a fixed mutation rate on some dynamic optimization problem instances. The proofs also offer some insights into conditions under which any time-variable mutation scheme is unlikely to be useful and into the relationships between the problem characteristics and algorithmic features (e.g., different mutation schemes).

0 citations0 reviewsNo code linkedOpen access