Feedback from peer review is essential to improve the quality of scientific articles. However, at present, many manuscripts do not receive sufficient external feedback for refinement before or during submission. Therefore, a system capable of providing detailed and professional feedback is crucial for enhancing research efficiency. In this paper, we have compiled the largest dataset of paper reviews to date by collecting historical open-access papers and their corresponding review comments and standardizing them using LLM. We then developed a multi-agent system that mimics real human review processes, based on LLMs. This system, named Agent Reviewers, includes the innovative introduction of multimodal reviewers to provide feedback on the visual elements of papers. Additionally, a shared memory pool that stores historical papers' metadata is preserved, which supplies reviewer agents with background knowledge from different fields. Our system is evaluated using ICLR 2024 papers and achieves superior performance compared to existing AI-based review systems. Comprehensive ablation studies further demonstrate the effectiveness of each module and agent in this system.

Video highlights detection (VHD) is an active research field in computer vision, aiming to locate the most user-appealing clips given raw video inputs. However, most VHD methods are based on the closed world assumption, i.e., a fixed number of highlight categories is defined in advance and all training data are available beforehand. Consequently, existing methods have poor scalability with respect to increasing highlight domains and training data. To address above issues, we propose a novel video highlights detection method named Global Prototype Encoding (GPE) to learn incrementally for adapting to new domains via parameterized prototypes. To facilitate this new research direction, we collect a finely annotated dataset termed LiveFood, including over 5,100 live gourmet videos that consist of four domains: ingredients, cooking, presentation, and eating. To the best of our knowledge, this is the first work to explore video highlights detection in the incremental learning setting, opening up new land to apply VHD for practical scenarios where both the concerned highlight domains and training data increase over time. We demonstrate the effectiveness of GPE through extensive experiments. Notably, GPE surpasses popular domain incremental learning methods on LiveFood, achieving significant mAP improvements on all domains. Concerning the classic datasets, GPE also yields comparable performance as previous arts. The code is available at: https://github.com/ForeverPs/IncrementalVHD_GPE.

We observe a high level of imbalance in the accuracy of different learned classes in the same old task for the first time. This intriguing phenomenon, discovered in replay-based Class Incremental Learning (CIL), highlights the imbalanced forgetting of learned classes, as their accuracy is similar before the occurrence of catastrophic forgetting. This discovery remains previously unidentified due to the reliance on average incremental accuracy as the measurement for CIL, which assumes that the accuracy of classes within the same task is similar. However, this assumption is invalid in the face of catastrophic forgetting. Further empirical studies indicate that this imbalanced forgetting is caused by conflicts in representation between semantically similar old and new classes. These conflicts are rooted in the data imbalance present in replay-based CIL methods. Building on these insights, we propose CLass-Aware Disentanglement (CLAD) as a means to predict the old classes that are more likely to be forgotten and enhance their accuracy. Importantly, CLAD can be seamlessly integrated into existing CIL methods. Extensive experiments demonstrate that CLAD consistently improves current replay-based methods, resulting in performance gains of up to 2.56%.

Continual learning strives to acquire knowledge across sequential tasks without forgetting previously assimilated knowledge. Current state-of-the-art methodologies utilize dynamic architectural strategies to increase the network capacity for new tasks. However, these approaches often suffer from a rapid growth in the number of parameters. While some methods introduce an additional network compression stage to address this, they tend to construct complex and hyperparameter-sensitive systems. In this work, we introduce a novel solution to this challenge by proposing Memory-Boosted transformer (MoBoo), instead of conventional architecture expansion and compression. Specifically, we design a memory-augmented attention mechanism by establishing a memory bank where the “key” and “value” linear projections are stored. This memory integration prompts the model to leverage previously learned knowledge, thereby enhancing stability during training at a marginal cost. The memory bank is lightweight and can be easily managed with a straightforward queue. Moreover, to increase the model’s plasticity, we design a memory-attentive aggregator, which leverages the cross-attention mechanism to adaptively summarize the image representation from the encoder output that has historical knowledge involved. Extensive experiments on challenging benchmarks demonstrate the effectiveness of our method. For example, on ImageNet-100 under 10 tasks, our method outperforms the current state-of-the-art methods by +3.74% in average accuracy and using fewer parameters.

Humans are proficient at continuously acquiring and integrating new knowledge. By contrast, deep models forget catastrophically, especially when tackling highly long task sequences. Inspired by the way our brains constantly rewrite and consolidate past recollections, we propose a novel Bilateral Memory Consolidation (BiMeCo) framework that focuses on enhancing memory interaction capabilities. Specifically, BiMeCo explicitly decouples model parameters into short-term memory module and long-term memory module, responsible for representation ability of the model and generalization over all learned tasks, respectively. BiMeCo encourages dynamic interactions between two memory modules by knowledge distillation and momentum-based updating for forming generic knowledge to prevent forgetting. The proposed BiMeCo is parameter-efficient and can be integrated into existing methods seamlessly. Extensive experiments on challenging benchmarks show that BiMeCo significantly improves the performance of existing continual learning methods. For example, combined with the state-of-the-art method CwD [55], BiMeCo brings in significant gains of around 2% to 6% while using 2x fewer parameters on CIFAR-100 under ResNet-18.

SUMMARY: Recent advances of long-term time-lapse microscopy have made it easy for researchers to quantify cell behavior and molecular dynamics at single-cell resolution. However, the lack of easy-to-use software tools optimized for customized research is still a major challenge for quantitatively understanding biological processes through microscopy images. Here, we present CellTracker, a highly integrated graphical user interface software, for automated cell segmentation and tracking of time-lapse microscopy images. It covers essential steps in image analysis including project management, image pre-processing, cell segmentation, cell tracking, manually correction and statistical analysis such as the quantification of cell size and fluorescence intensity, etc. Furthermore, CellTracker provides an annotation tool and supports model training from scratch, thus proposing a flexible and scalable solution for customized dataset analysis. AVAILABILITY AND IMPLEMENTATION: CellTracker is an open-source software under the GPL-3.0 license. It is implemented in Python and provides an easy-to-use graphical user interface. The source code, instruction manual and demos can be found at https://github.com/WangLabTHU/CellTracker. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Customizing the precision of data can provide attractive trade-offs between accuracy and hardware resources. Custom hardware and FPGA designs allow bit-level control over precision, but software is typically limited by the range of types supported by the underlying processor. We propose a new form of vector computing aimed at arrays of custom-precision data on general-purpose processors with SIMD extensions. We represent these vectors in bitslice format and use bitwise instructions to build arithmetic operators that operate on the customized bit precision. We construct a domain-specific code generator that builds bit-level customizable floating-point and integer operators for our vector types. Using a hardware circuit optimization tool we optimize our logical expressions, and synthesize fast software arithmetic operators for bitslice vector types. We evaluate the resulting code and find that advanced logic optimization significantly improves performance. Experiments on a platform with Intel AVX2 SIMD extensions show that this approach is efficient for vectors of low-precision custom floating-point types, while providing arbitrary bit precision.

Nested thread-level parallelism (TLP) is pervasive in real applications. For example, 75% (14 out of 19) of the applications in the Rodinia benchmark for heterogeneous accelerators contain kernels with nested thread-level parallelism. Efficiently mapping the enclosed nested parallelism to the GPU threads in the C-to-CUDA compilation (OpenACC in this paper) is becoming more and more important. This mapping problem is two folds: suitable execution models and efficient mapping strategies of the nested parallelism.

Stencil computations are core of wide range of scientific and engineering applications. A lot of efforts have been put into improving efficiency of stencil calculations on different platforms, but unfortunately it is not easy to reuse. In this paper we present a PAttern-Driven Stencil compiler-based tool and a simple tuning system to reuse those well optimized methods and codes. We also suggest extensions to OpenMP, depicting high-level data structures in order to facilitate recognition of various stencil computation patterns. The PADS allows programmers to rewrite kernel of stencils or reuse source-to-source translator outputs as optimized stencil template codes with related tuning parameters, In addition, PADS consists of a OpenMP to CUDA translator and code generator using optimized template codes. It also obtains architecture-specific parameters to tune stencils across different GPU platforms. To demonstrate our system flexibility and performance portability, we illustrate four different stencil computations, Laplacian operator with Jacobi iterative method, divergence operator, 3 dimension 25 point stencil and a 2D heat equation using ADI method with periodic boundary conditions. PADS succeeds in generating all these four stencil codes using different optimization strategies and delivers a promising performance improvement.

The wide application of General Purpose Graphic Processing Units (GPGPUs) results in large manual efforts on porting and optimizing algorithms on them. However, most existing automatic ways of generating GPGPU code fail to conduct optimization strategies regarding a specific computation and to reuse constantly evolving manual optimizations. In this paper, we present a computation pattern driven approach for computation-specific GPGPU code generation and optimization, which in turn reuses manual optimizations to a certain extent. We suggest language extensions to OpenMP, high-level data structure attributes, in order to assist the process of computation pattern matching and to help give users intuitive performance tuning parameters in the view of data structure attributes. We illustrate the feasibility of this approach through three important computation dwarfs, which are dense matrix, sparse matrix, and structured mesh computation in scientific computing. We also build a prototype OpenMP-to-CUDA translator that consists of computation pattern recognition and code template instantiation. The experimental results demonstrate the performance benefits of computation pattern driven method. To our best knowledge, it is the first work on reusing manual optimizations for GPGPUs with computation pattern driven approach.