• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.3
• /
• pp.837-852
• /
• 2021

Recurrent neural network (RNN) architectures have been used for language modeling (LM) tasks that require learning long-range word or character sequences. However, the RNN architecture is still suffered from unstable gradients on long-range sequences. To address the issue of long-range sequences, an attention mechanism has been used, showing state-of-the-art (SOTA) performance in all LM tasks. A differentiable neural computer (DNC) is a deep learning architecture using an attention mechanism. The DNC architecture is a neural network augmented with a content-addressable external memory. However, in the write operation, some information unrelated to the input word remains in memory. Moreover, DNCs have been found to perform poorly with low numbers of weight parameters. Therefore, we propose a robust memory deallocation method using a limited retention vector. The limited retention vector determines whether the network increases or decreases its usage of information in external memory according to a threshold. We experimentally evaluate the robustness of a DNC implementing the proposed approach according to the size of the controller and external memory on the enwik8 LM task. When we decreased the number of weight parameters by 32.47%, the proposed DNC showed a low bits-per-character (BPC) degradation of 4.30%, demonstrating the effectiveness of our approach in language modeling tasks.

• Proceedings of the IEEK Conference
• /
• 2003.07d
• /
• pp.1335-1338
• /
• 2003

Fixed-size memory allocation is one of the most promising way to avoid external fragmentation in dynamic memory allocation problem. This paper presents an experimental result of applying the fixed- size memory allocation strategy to Java virtual machine for embedded system. The result says that although this strategy induces another memory utilization problem caused by internal fragmentation, the effect is not very considerable and this strategy is well-suited for embedded Java system. The experiment has been performed in a real embedded Java system called the simpleRTJ.

• Proceedings of the Korean Information Science Society Conference
• /
• 2002.10d
• /
• pp.655-657
• /
• 2002

메모리 반납(deallocation) 명령어는 프로그램에게 할당된 힙 셀(heap cell)을 반납하는 명령어로 힙 사용량을 낮추어 주지만, 잘못된 반납으로 인해 심각한 오류를 일으킬 수 있다. 본 논문에서는 재귀적인 자료구조(recursive data structure),를 안전하게 반납하는 명령어를 삽입하는 알고리즘을 제시한다. 메모리의 모양새를 분석하고 나중에 쓰이지 않을 힙 셀들을 추정하여 반납 명령어를 삽입한다. 분석시 요약 수준을 적절히 조절함으로써 빠르면서도 정확하게 분석한다. 또한, 실행시간에 부가적인 정보를 전달하여 일찍 힙 셀을 반납할 수 있도록 한다. 제시한 알고리즘으로 메모리 반납을 하지 않는 프로그램에 반납 명령어를 삽입하여 전체 메모리 할당량의 5.2-98.7%를 반납할 수 있었다.

• KIPS Transactions on Computer and Communication Systems
• /
• v.4 no.9
• /
• pp.327-336
• /
• 2015

Mobile game servers usually execute frequent dynamic memory allocation for generating the buffers that deal with clients requests. It causes to deteriorate the performance of game servers since it increases system workload and memory fragmentation. In this paper, we propose fixed-sized memory pool management method. Memory pool for the proposed method has a sequential memory structure based on circular linked list data structure. It solves memory fragmentation problem and saves time for searching the memory blocks which are required for memory allocation and deallocation. We showed the efficiency of the proposed method by evaluating the performance of dynamic memory allocation, through the proposed method and the memory pool management method based on boost open source library.