8 research outputs found
Exploiting procedure level locality to reduce instruction cache misses
No full textHigh instruction fetch bandwidth is essential for high performance in today’s wide-issue outof-order processors. Instruction caches must provide a low miss rate as well as low latency. We introduce Procedure Level Relocation, a class of dynamic feedback-directed optimizations that substantially reduce the instruction cache miss rate by exploiting the temporal locality of procedure usage. Based on the observation that half of all procedures executed are at most 128 bytes in length, we present a Small Procedure Cache, a small and fast explicitly managed memory for storing small procedures. We show that Procedure Level Relocation into a Small Procedure Cache reduces the instruction cache miss rate by an average of 15%Technical report DCS-TR-53
Working Sets at Function Level
No full textThe trend in computer architecture is that of an increasing gap between rapidly increasing processor speeds and the relatively slower memory subsystems. Program locality has played a crucial role in engineering economically viable computer systems. While program locality has been studied and exploited at different levels of memory hierarchy, including the cache, block and page levels, there has been little or no effort to study locality at the level of functions and to leverage it.
In this paper we show that there is a strong correlation between current function references and those in the immediate future, whereas, those in the distant future tend to be more or less uncorrelated. In particular we show that an average working set (WS) size of 20% of the program size in terms of number of functions is sufficient, in all cases, to ensure a miss rate of less than 2%. Moreover, we also observe that in most cases a WS of half that size is good enough to ensure similar low miss rates. We also compare two different kinds of WSs -- fixed size maintained by LRU and variable size based on a WS parameter -- and show that the variable size WS consistently performs better. Finally we present a novel way of exploiting the locality at function level.Technical report DCS-TR-48
A Study of Program Behavior to Establish Temporal Locality at the Function Level
No full textThe trend in computer architecture is that processor speeds are increasing rapidly compared to memory access times and the relatively stagnant disk speed. Computer software, on the other hand is characterized by growing program sizes and sophisticated functionality. The combination of these factors has resulted in a processor memory bottleneck, which is worsening with time. While program behavior has been studied at page level and cache level and the locality at page, cache and block levels has been exploited, there has been comparatively much lesser amount of work to exploit locality at the level of functions, and no prior work to study program behavior at this level.
In this paper we show that there is a considerable amount of temporal locality at the level of functions. In particular we show that a working set of functions containing 40% of all the programs functions results in a miss rate of less than 2%. Moreover, we observe that, in almost all cases, even working sets having half as many functions result in similar low miss rates. Our experiments indicate that program execution is characterized by a working set of functions which changes with time and a thrashing like phenomenon results when the function footprint is not resident in memory.Technical report DCS-TR-47
Enhancing Educational Assessment Efficiency: An AIbased Approach to Automated Examination Evaluation
No full textManually grading subjective papers is a difficult and time-consuming task. Lack of comprehension and acceptanceof the results is one of the biggest obstacles to employing artificial intelligence (AI) to analyse subjective articles.There have been several attempts to grade responses from pupils using computer science. However, most of thework uses specific words or traditional counts to achieve this. Furthermore, verified data sets are not enough.Using a range of natural language processing, machine learning, and toolkits, including Wordnet, Word2vec, wordmover's distance (WMD), cosine comparison, multinomial naive bayes (MNB), and others, this study presents anovel approach for automatically analysing descriptive responses and TF-IDF, or term frequency-inversedocument frequency. Answers are evaluated using keywords and solution statements, and grades are predictedusing a machine learning algorithm. Overall, the findings show that WMD and cosine are more comparable. Themachine learning system may be used independently once it has undergone the required training. Trial and errorresults 
Utilizing the Ethereum blockchain for retrieving and archiving augmented reality surgical navigation data.
No full textAIM: Conventional techniques to share and archive spinal imaging data raise issues with trust and security, with novel approaches being more greatly considered. Ethereum smart contracts present one such novel approach. Ethereum is an open-source platform that allows for the use of smart contracts. Smart contracts are packages of code that are self-executing and reside in the Ethereum state, defining conditions for programmed transactions. Though powerful, limited attempts have been made to showcase the clinical utility of such technologies, especially in the pre- and post-operative imaging arenas. Herein, we therefore aim to propose a proof-of-concept smart contract that stores intraoperative three-dimensional (3D) augmented reality surgical navigation (ARSN) data and was tested on a private, proof-of-authority network. To the author\u27s best knowledge, the present study represents a first-use case of the Interplanetary File Storage protocol for storing and retrieving spine imaging smart contracts.
METHODS: The content identifier hashes were stored inside the smart contracts while the interplanetary file system (IPFS) was used to efficiently store the image files. Insertion was achieved with four storage mappings, one for each of the following: fictitious patient data, specific diagnosis, patient identity document (ID), and Gertzbein grade. Inserted patient observations were then queried with wildcards. Insertion and retrieval times for different record volumes were collected.
RESULTS: It took 276 milliseconds to insert 50 records and 713 milliseconds to insert 350 records. Inserting 50 records required 934 Megabyte (MB) of memory per insertion with patient data and imaging, while inserting 350 records required almost the same amount of memory per insertion. In a database of 350 records, the retrieval function needs about 1,026 MB to query a record with all three fields left blank, but only 970 MB to obtain the same observation from a database of 50 records.
CONCLUSIONS: The concept presented in this study exemplifies the clinical utility of smart contracts and off-chain data storage for efficient retrieval/insertion of ARSN data
Apatone® induces endometrioid ovarian carcinoma (MDAH 2774) cells to undergo karyolysis and cell death by autoschizis: A potent and safe anticancer treatment
No full textOvarian cancers are still the most lethal gynecologic malignancy. As a novel strategy against this poor outcome cytotoxic alterations induced by a pro-oxidant treatment on human ovarian endometrioid carcinoma (MDAH 2774) cells are revisited by using light, scanning and transmission electron microscopy. A series of sequential and concomitant cellular and organelle injuries induced by ascorbate: menadione combination (VC: VK3) or Apatone® is emphasized. This adjuvant or treatment is able to kill majority of these tumor cells through ‘autoschizic cell death’, a mode of cell death different than apoptosis. Autoschizic cell death is significant after a short period of treatment to decrease the ovarian tumor cell population through induced injuries that proceed from membranes to most organelles: karyolysis with nucleolar segregation and fragmentation, autophagy of mitochondria, lysosome and other organelles as well as cytoskeletal defects. The cytoskeletal damages are evidenced by morphology changes that included auto- or self-excised pieces of cytoplasm lacking organelles apparently facilitated by grouping of vacuolated endoplasm. These results obtained against this endometrioid ovary cell line are comforted by other studies using Apatone® against other carcinomas in vitro and in vivo. Altogether these reports support Apatone® as a new drug that can favorably be used as a novel potent, safe, and inexpensive clinical adjuvant or treatment against ovarian cancers
New light for one-world approach toward safe and effective control of animal diseases and insect vectors from leishmaniac perspectives
No full text© 2016 The Author(s). Light is known to excite photosensitizers (PS) to produce cytotoxic reactive oxygen species (ROS) in the presence of oxygen. This modality is attractive for designing control measures against animal diseases and pests. Many PS have a proven safety record. Also, the ROS cytotoxicity selects no resistant mutants, unlike other drugs and pesticides. Photodynamic therapy (PDT) refers to the use of PS as light activable tumoricides, microbicides and pesticides in medicine and agriculture. Here we describe photodynamic vaccination (PDV) that uses PDT-inactivation of parasites, i.e. Leishmania as whole-cell vaccines against leishmaniasis, and as a universal carrier to deliver transgenic add-on vaccines against other infectious and malignant diseases. The efficacy of Leishmania for vaccine delivery makes use of their inherent attributes to parasitize antigen (vaccine)-presenting cells. Inactivation of Leishmania by PDT provides safety for their use. This is accomplished in two different ways: (i) chemical engineering of PS to enhance their uptake, e.g. Si-phthalocyanines; and (ii) transgenic approach to render Leishmania inducible for porphyrinogenesis. Three different schemes of Leishmania-based PDV are presented diagrammatically to depict the cellular events resulting in cell-mediated immunity, as seen experimentally against leishmaniasis and Leishmania-delivered antigen in vitro and in vivo. Safety versus efficacy evaluations are under way for PDT-inactivated Leishmania, including those further processed to facilitate their storage and transport. Leishmania transfected to express cancer and viral vaccine candidates are being prepared accordingly for experimental trials. We have begun to examine PS-mediated photodynamic insecticides (PDI). Mosquito cells take up rose bengal/cyanosine, rendering them light-sensitive to undergo disintegration in vitro, thereby providing a cellular basis for the larvicidal activity seen by the same treatments. Ineffectiveness of phthalocyanines and porphyrins for PDI underscores its requirement for different PS. Differential uptake of PS by insect versus other cells to account for this difference is under study. The ongoing work is patterned after the one-world approach by enlisting the participation of experts in medicinal chemistry, cell/molecular biology, immunology, parasitology, entomology, cancer research, tropical medicine and veterinary medicine. The availability of multidisciplinary expertise is indispensable for implementation of the necessary studies to move the project toward product development
Global PIQA: Evaluating Physical Commonsense Reasoning Across 100+ Languages and Cultures
No full textTo date, there exist almost no culturally-specific evaluation benchmarks for large language models (LLMs) that cover a large number of languages and cultures. In this paper, we present Global PIQA, a participatory commonsense reasoning benchmark for over 100 languages, constructed by hand by 335 researchers from 65 countries around the world. The 116 language varieties in Global PIQA cover five continents, 14 language families, and 23 writing systems. In the non-parallel split of Global PIQA, over 50% of examples reference local foods, customs, traditions, or other culturally-specific elements. We find that state-of-the-art LLMs perform well on Global PIQA in aggregate, but they exhibit weaker performance in lower-resource languages (up to a 37% accuracy gap, despite random chance at 50%). Open models generally perform worse than proprietary models. Global PIQA highlights that in many languages and cultures, everyday knowledge remains an area for improvement, alongside more widely-discussed capabilities such as complex reasoning and expert knowledge. Beyond its uses for LLM evaluation, we hope that Global PIQA provides a glimpse into the wide diversity of cultures in which human language is embedded.See §A for author list. Global PIQA would not be possible without the efforts of all of the authors. Wealso thank several anonymous contributors who preferred not to be authors on this paper. The research of Yolanda Xavier is supported by Portuguese national funding through the FCT– Portuguese Foundation for Science and Technology, I.P. as part of the project UID/3213/2025– Linguistics Research Centre of NOVA University Lisbon (CLUNL) and by the Doctoral Grant (FCT PhD grant) number 2022.13977.BD from the same funder. Group 0025 is supported by the following grants: CLARIN-PL (POIR.04.02.00-00C002/19, FENG.02.04-IP.040004/24, 2024/WK/01), DARIAH-PL (POIR.04.02.00-00-D006/20, KPOD.01.18-IW.03-0013/23). Annika Simonsen was funded by the European Commission under grant agreement no. 101135671. CEB has been partially funded by the German ministry for education and research (BMBF) through the TRAILS project (grant number 01IW24005). Group 0070 is supported by funding from King Abdullah University of Science and Technology (KAUST)- Center of Excellence for Generative AI, under award number 5940. Group 0079 would like to thank Mr. Sudhir R. Narayana for help with correction and verification of items in their dataset. Sina Ahmadi gratefully acknowledges support from the University of Zurich (UZH) Postdoc Grant (reference number 269093). Group 0133 would like to thank the MbazaNLP community, including students from the University of Rwanda, School of Art and Languages. We would also like to thank Yonatan Bisk for useful insights into the original PIQA dataset
