AYJ : Microservices - when and how to use them

Tutor: Antti Ylä-Jääski (antti.yla-jaaski@aalto.fi)

Microservice architecture is a modern approach to software system design in which the system's functionality is divided into small independent units. Microservice systems differ from a more traditional monolithic system in many ways some of which are unexpected. The cost of a migration from a monolithic system to a system based on microservices is often substantial so this decision needs to be carefully evaluated. An increasing number of companies (e.g., Amazon, Netflix, LinkedIn) are moving towards dismantling their existing monolithic applications in favor of distributed microservice systems. In this project, you will discuss the benefits and drawbacks of adopting microservice architecture in comparison to monolithic architecture. Depending on your interests, possible viewpoints on this generic topic are also for example security challenges, scalability, serverless computing, service mesh, and big data platforms.

References:

  • Nicola Dragoni, Saverio Giallorenzo, Alberto Lluch Lafuente, Manuel Mazzara, Fabrizio Montesi, Ruslan Mustafin, and Larisa Safina.Microservices: Yesterday, Today, and Tomorrow, pages 195–216. Springer International Publishing, Cham, 2017.
  •  Pooyan Jamshidi, Claus Pahl, Nabor C. Mendon.a, James Lewis, and KStefan Tilkov. Microservices: The journey so far and challenges ahead. IEEE SOFTWARE, 35(3):24 – 35, 2018.

AI : Predicting Depression through Digital Phenotyping
Tutor: Arsi Ikäheimonen (arsi.ikaheimonen@aalto.fi)

In the field of digital mental health, particularly regarding depression, the identification of early warning signals and the increased risk for relapse would be crucial for timely intervention and treatment. The increasingly common use of smartphones and wearable devices has led to the creation of a new area of study known as digital phenotyping. This approach leverages digital traces, the byproduct of our interactions with these devices, to monitor mental health. These traces provide a rich, unobtrusive, continuous data stream that can reveal behavioral patterns indicative of mental health states. In this project, you will conduct a focused literature review on the recent advancements in digital phenotyping as a tool for predicting depression relapses and identifying associated early warning signs. You will review important literature to find and assess effective statistical methods and machine and deep learning strategies, focusing on how digital traces—like phone usage patterns, social media interactions, and location data from wearables—can be harnessed to construct predictive models. Subsequently, you will synthesize these findings to present a comprehensive overview of the potential applications in mental health monitoring.


References:

  • Digital Phenotyping: https://doi.org/10.1038/s41386-020-0771-3 
  • Digital Phenotyping for Monitoring Mental Disorders: https://doi.org:10.2196/46778

SS1 : Self-help applications for mental disorders

Tutor: Sanna Suoranta (sanna.suoranta@aalto.fi)

Nowadays in Finland, mental disorders are bigger reason to early retirement than physical disorders. Many doctors or therapist use mobile applications to follow state of their patients but there is also self-help applications. What technical issues do these
digital mobile health applications face? Do they have common ways to engage their users and do they really help people suffering (mild) mental disorders? (Add your own questions...)

References:

  • Nicholas A. Livingston, Rebecca Shingleton, Meagan E. Heilman and Deborah Brief. 2019. Self-help Smartphone Applications for Alcohol Use, PTSD, Anxiety, and Depression: Addressing the New Research-Practice Gap. Journal of Technology in Behavioral Science. Vol 4. pages 139-151https://doi.org/10.1007/s41347-019-00099-6


  • Jiaqi Song, Ronghuan Jiang, Nan Chen, Wei Qu, Dan Liu, Meng Zhang, Hongzhen Fan, Yanli Zhao, and Shuping Tan. 2021. Self-help cognitive behavioral therapy application for COVID-19-related mental health problems: A longitudinal trial. Asian Journal of Psychiatry. Vol 60, June 2021.https://doi.org/10.1016/j.ajp.2021.102656


  • Arfan Ahmed, Nashva Ali, Anna Giannicchi, Alaa A Abd-alrazaq, Mohamed Ali Siddig Ahmed, Sarah Aziz, and Mowafa Househ. 2021.Mobile applications for mental health self-care: A scoping review.
    Computer Methods and Programs in Biomedicine Update
    Volume 1, 2021.
    https://doi.org/10.1016/j.cmpbup.2021.100041


SS2 : It is January, so I start now a new habit of.. 

Tutor: Sanna Suoranta (sanna.suoranta@aalto.fi)

Again, it is beginning of a new year,  January, Monday, or another reason to start a new habit that promotes good healthy life! When population gets older and there are less workers, societies triest to get their citizens to keep themselves in as good shape as possible. To motivate themselves, people uses self-help applications or wearable devices to follow their habits, but devices can also be given by medical doctors to follow the state of a patient.  What are the essential things for making good mobile health applications?

References:

  • Preethi R Sama, Zubin J Eapen, Kevin P Weinfurt, Bimal R Shah, and Kevin A Schulman. 2014. An Evaluation of Mobile Health Application ToolsJMIR Mhealth Uhealth 2014;2(2):e19
    doi: 10.2196/mhealth.3088



  • Daiana Biduski, Ericles Andrei Bellei, João Pedro Mazuco Rodriguez, Luciana Aparecida Martinez Zaina, and Ana Carolina Bertoletti De Marchi. 2020.Assessing long-term user experience on a mobile health application through an in-app embedded conversation-based questionnaire
    Computers in Human Behavior, Volume 104, March 2020
    https://doi.org/10.1016/j.chb.2019.106169


  • Olga Vl. Bitkina, Hyun K. Kim, and Jaehyun Park  Usability and user experience of medical devices: An overview of the current state, analysis methodologies, and future challenges. 2020. International Journal of Industrial Ergonomics, Volume 76, March 2020https://doi.org/10.1016/j.ergon.2020.102932

  • Jiaqi Song, Ronghuan Jiang, Nan Chen, Wei Qu, Dan Liu, Meng Zhang, Hongzhen Fan, Yanli Zhao, and Shuping Tan. 2021. Self-help cognitive behavioral therapy application for COVID-19-related mental health problems: A longitudinal trial. Asian Journal of Psychiatry. Vol 60, June 2021.https://doi.org/10.1016/j.ajp.2021.102656

SS3 : Games for cognitive abilities of elderly people

Tutor: Sanna Suoranta (sanna.suoranta@aalto.fi)

According to research, the best way to prevent cognitive abilities to decline due aging, is to keep learning new skills (dancing is best way but e.g. learning new languages is good, too). One way to engage learning is to gamify it. Games can be used both entretaining purposes but also for improving cognitive abilities after, for example, strokes. What is essential in non-entretaining games targeted for elderly people?

References:

  • Sabrina Oppl and Christian Stary.Game-playing as an effective learning resource for elderly people: encouraging experiential adoption of touchscreen technologies.
    Universal Access in the Information Society 19, 295–310 (2020). Published 3.11.2018.
    https://doi.org/10.1007/s10209-018-0638-0

  • Abd-alrazaq A, Alajlani M, Alhuwail D, Toro CT, Giannicchi A, Ahmed A, Makhlouf A, Househ MThe Effectiveness and Safety of Serious Games for Improving Cognitive Abilities Among Elderly People With Cognitive Impairment: Systematic Review and Meta-Analysis. JMIR Serious Games 2022;10(1):e34592
    doi: 10.2196/34592


  • Antonio Rienzo and Claudio Cubillos.Playability and Player Experience in Digital Games for Elderly: A Systematic Literature Review. 2020. Sensors 2020, 20(14)
    https://doi.org/10.3390/s20143958



JH : Predicting the energy consumption of cloud applications

Tutor: Jaakko Harjuhahto (jaakko.harjuhahto@aalto.fi)

Deploying virtual machines and containerized applications are central activities for operating cloud systems. The target machine where the new application instance is placed must, at minimum, have free resources such as CPU and RAM to run the application. More sophisticated placement algorithms will consider more factors than satisfying these minimum requirements. Typically the placement decision is an optimization problem that must balance resource availability, quality of service and the expected cost of running the application. Energy efficiency [1] is a major component of the expected costs, and of increasing importance for the industry to meet sustainability goals. However, predicting the energy consumption of cloud applications accurately and precisely remains a challenge. For this topic, review recent literature on the state of the art for methods to predict cloud application resource consumption.

References:

  •  Rajeev Muralidhar, Renata Borovica-Gajic, Rajkumar Buyya. "Energy Efficient Computing Systems: Architectures, Abstractions and Modeling to Techniques and Standards". ACM Computing Surveys, September 2022. https://doi.org/10.1145/3511094


JLMN : CDN attacks

Tutor: Jose Luis Martin Navarro (jose.martinnavarro@aalto.fi)

Content Delivery Networks (CDNs) are an essential Internet infrastructure that improves the performance and scalability of content requests such as webpages and media. They work as a geographically distributed proxy platform, caching and forwarding content on a massive scale. Many CDN providers include Distributed Denial-of-Service (DDoS) protection as an additional feature of using their services, usually through tools such as Web application firewalls (WAF). Several studies have tried to develop a general model for CDN attacks [1, 2] . However, the CDN landscape is not static. New variants of attacks emerge every year, and vendors fix and develop different countermeasures. In this project, we ask students to conduct a literature review on the latest DDoS attacks [3, 4, 5, 6, 7] and validate the results with an ethical approach. Since it is a broad area, each student will decide an area to focus on after starting to work on the topic (cache attacks, amplification, HTTP vulnerabilities, countermeasures, …). With this understanding, we aim to identify new potential threats for CDNs.

References:

  • M. Ghaznavi, E. Jalalpour, M. A. Salahuddin, R. Boutaba, D. Migault and S. Preda, "Content Delivery Network Security: A Survey," in *IEEECommunications Surveys & Tutorials*, vol. 23, no. 4, pp. 2166-2190,Fourthquarter 2021, doi: 10.1109/COMST.2021.3093492. 
  • - Sharafaldin, Iman, Arash Habibi Lashkari, Saqib Hakak, and Ali A.Ghorbani. "Developing realistic distributed denial of service (DDoS)attack dataset and taxonomy." In *2019 International Carnahan Conference on Security Technology (ICCST)*, pp. 1-8. IEEE, 2019. - Li, W., Shen, K., Guo, R., Liu, B., Zhang, J., Duan, H., ... & Wang, Y. (2020, June).
  •  CDN backfired: amplification attacks based on http range requests. In *2020 50th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN)* (pp. 14-25). IEEE. 
  • - Guo, Run, Jianjun Chen, Yihang Wang, Keran Mu, Baojun Liu, Xiang Li,Chao Zhang, Haixin Duan, and Jianping Wu. "Temporal {CDN-Convex} Lens: A {CDN-Assisted} Practical Pulsing {DDoS} Attack." In *32nd USENIX Security Symposium (USENIX Security 23)*, pp. 6185-6202. 2023 
  • - Li, Zihao, and Weizhi Meng. "Mind the amplification: cracking content delivery networks via DDoS attacks." In *Wireless Algorithms, Systems, and Applications: 16th International Conference,WASA 2021, Nanjing, China, June 25–27, 2021, Proceedings, Part II 16*, pp. 186-197. Springer International Publishing, 2021
  •  - Guo, Run, Weizhong Li, Baojun Liu, Shuang Hao, Jia Zhang, Haixin Duan, Kaiwen Sheng, Jianjun Chen, and Ying Liu. "CDN Judo: Breaking the CDN DoS Protection with Itself." In *NDSS*. 2020. 
  • - Shen, Kaiwen, Jianyu Lu, Yaru Yang, Jianjun Chen, Mingming Zhang, Haixin Duan, Jia Zhang, and Xiaofeng Zheng. "HDiff: A Semi-automatic Framework for Discovering Semantic Gap Attack in HTTP Implementations." In *2022 52nd Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN)*, pp. 1-13. IEEE, 2022.


JV1 : Signals - new standard for state management in the web?

Tutor: Juho Vepsäläinen (juho.vepsalainen@aalto.fi)

The purpose of this study is to look into a state management solution called signals and how it differs from earlier options. Lately signals have been implemented for several frameworks/libraries, such as Preact, Solid, and Qwik and more implementations exist..

References:

  •  https://www.solidjs.com/ 
  • https://preactjs.com/guide/v10/signals/ 
  • https://qwik.builder.io/docs/components/state/

JV2 : State orchestration on the web

Tutor: Juho Vepsäläinen (juho.vepsalainen@aalto.fi)

State machines and charts form an alternative way to orchestrate state for web applications. The purpose of the study would be to understand the potential of the approach better and capture the current state of art.

References:

  •  https://xstate.js.org/

JV3 : Utility first CSS for the web
Tutor: Juho Vepsäläinen (juho.vepsalainen@aalto.fi)

Utility first styling solutions have become popular during the past few years. Tailwind is perhaps the most known example but several others exist. The purpose of the study would be to understand why the approach grew in popularity and highlight its opportunities and challenges.

References:

  • https://tailwindcss.com/ 
  • https://unocss.dev/


MS : Neural video coding
Tutor: Matti Siekkinen (matti.siekkinen@aalto.fi)

Neural video compression has gained much interest over the last years. This topic focuses specifically on end-to-end neural video coding as opposed to enhancing traditional video coding methods with learning based components. The student will study the state of the art and recent advances in neural video coding from a chosen perspective. Practical experimentation is also possible.


References:

  • Dong Liu, Yue Li, Jianping Lin, Houqiang Li, and Feng Wu. 2020. Deep Learning-Based Video Coding: A Review and a Case Study. ACM Comput. Surv. 53, 1, Article 11 (January 2021), 35 pages. 
  • https://doi.org/10.1145/3368405 Chen, H., He, B., Wang, H., Ren, Y., Lim, S. N., & Shrivastava, A. (2021). 
  • Nerv: Neural representations for videos. Advances in Neural Information Processing Systems, 34, 21557-21568. 
  •  van Rozendaal, T., Singhal, T., Le, H., Sautiere, G., Said, A., Buska, K., ... & Wiggers, A. (2024). MobileNVC: Real-time 1080p
  •  Neural Video Compression on a Mobile Device. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 4323-4333).

MK1 : Metrics for machine learning robustness against adversarial attacks
Tutor: Mikko Kiviharju (mikko.kiviharju@aalto.fi)

Some machine learning types are vulnerable to different data-level attacks, such as data poisoning, backdooring or evasion. There are, however, methods to make implementations more robust. The robustness itself can be measured in multiple ways (trivially e.g. by attack success rate), but this varies by the exact NN type and attack model. This topic can be approached either as a survey, by identifying and mapping different robustness metrics with an associated comparative study; or by selecting one or two specific metrics and investigating them in a detailed manner.


References:

  • BSI, Federal Office for Information Security: "Security of AI-Systems:Fundamentals - Adversarial Deep Learning" https://www.bsi.bund.de/SharedDocs/Downloads/EN/BSI/KI/Security-of-AI-systems_fundamentals.html 
  •  NIST: "Adversarial Machine Learning A Taxonomy and Terminology of Attacks and Mitigations", https://doi.org/10.6028/NIST.AI.100-2e2023.ipd 
  •  Formin/Kansallinen turvallisuusviranomainen: "KATAKRI 2020: Tietoturvallisuuden auditointityökalu viranomaisille", https://um.fi/katakri-tietoturvallisuuden-auditointityokalu-viranomaisille

MK2 : Measuring cybersecurity risk of industrial control / OT systems
Tutor: Mikko Kiviharju (mikko.kiviharju@aalto.fi)

Industrial control systems (IACS) or more widely operational technology (OT) presents an ever-increasing attack surface for cyber criminals and state actors. To remedy the situation, there are standardized guidelines to help organizations build their cybersecurity risk management systems. One of the most popular ones is the IEC-62443 standards series. The series contains methods to classify assets according to their sensitivity, controls by their capability and threats by their magnitude. The student should familiarize itself with this risk valuation metrics and either apply it to an existing (or hypothetical but plausible) system, or make a survey or public applications of the framework as a comparative study.


References:

  • IEC-62443 standard series


SR1 : Designing distributed mechanisms for local problems 
Tutor: Sara Ranjbaran (sara.ranjbaran@aalto.fi)

In this work we study a local version of the classical Gale–Shapley algorithm  for the stable marriage problem.


References:

  • Floréen, Patrik, et al. "Almost stable matchings by truncating the Gale–Shapley algorithm." Algorithmica 58 (2010): 102-118

SR2 :User-Provided-Infrastructure at the Edge
Tutor: Sara Ranjbaran (sara.ranjbaran@aalto.fi)


Today, we are witnessing an ever increasing demand for ubiquitous user connectivity and computation, and at the same time, the surging of advanced handheld devices, with enhanced storage and computing capabilities. These advanced devices can not only satisfy the needs of their owners but also act as a host to provide related services to users nearby. Recently, the concept of Fog-Provided-Infrastructure (FPI), where heterogeneous devices pool their resources to match wide-ranging user requirements has been attracting increasing attention.

References:

  • I. Farris, L. Militano, M. Nitti, L. Atzori, and A. Iera, “Mifaas: A mobile-iot-federation-as-a-service model for dynamic cooperation of iot cloud providers,” Future Generation Computer Systems, vol. 70, pp. 126–137, 2017.


NHK : Breathing Life into Virtual Characters with Embodied Intelligence
Tutor: Nam Hee Kim (namhee.kim@aalto.fi)

Team ARAM presents: Breathing Life into Virtual Characters with Embodied Intelligence Character animation remains an extremely attractive testbed for artificial intelligence techniques with many exciting downstream applications such as robotics, films, games, and many more. In this seminar project, we will investigate together the frontiers of character animation and embodied intelligence research, examining the use of generative AI (e.g. transformers, diffusion) and deep reinforcement learning. Students will regularly read and discuss most recent SIGGRAPH / CVPR / ICLR / ICML papers as part of the learning journey. The selected candidates will regularly participate in paper reading and project meetings hosted by Team ARAM (Animation, Robotics, and Machine Learning), a special research interest group focusing on building a culture of peer advising and disciplined research collaboration in method-centric topics. While it is possible to do literature review with this project, we will put priority on driven and motivated candidates who are willing to get hands-on experience with machine learning experiments, with the aim of producing a workshop-tier paper by the end of the project.

References:

  • Team ARAM meeting slides:
  • https://docs.google.com/presentation/d/1H3xe6H3UJITgeZ03Bj3Auf3Rfl_fStBc2pmvPDx6nds/edit?usp=sharing
  •  Exemplary works in this domain: 
  •  https://moconvq.github.io/ MoConVQ: Unified Physics-Based Motion Control via Scalable Discrete Representations 
  •  https://vcai.mpi-inf.mpg.de/projects/FatiguedMovements/ Discovering Fatigued Movements for Virtual Character Animation 
  •  https://nvlabs.github.io/PhysDiff/ PhysDiff: Physics-Guided Human Motion Diffusion Model 
  • https://azadis.github.io/make-an-animation/ Make-An-Animation: Large-Scale Text-conditional 3D Human Motion Generation (ICCV 2023) 
  •  https://tencent-roboticsx.github.io/NCP/ Neural Categorical Priors for Physics-Based Character Control 
  • https://www.cs.ubc.ca/~van/papers/2022-SIGGRAPH-juggle/index.html Learning Soccer Juggling Skills with Layer-wise Mixture-of-Experts ... 
  • and many more! (email namhee.kim@aalto.fi for more references)


PCH : Theory of Clustering: Algorithms and Optimization
Tutor: Parinya Chalermsook (parinya.chalermsook@aalto.fi)

Clustering is among the most important task in computer science that arises in algorithms, machine learning, and optimization. The goal of this topic is to read recent theoretical developments on clustering. This topic will focus less on applications and more on algorithmic and optimization challenges of the tasks


References:

  • A good starting point is Chapter 7 in this book: https://www.cs.cornell.edu/jeh/book.pdf


RWFL : Doubly Efficient Private Information Retrieval
Tutor: Russell W. F.  Lai(russell.lai@aalto.fi)

A private information retrieval (PIR) protocol is a two-round client-server protocol which allows the client to learn the value of an entry of the database held by the server, without the server knowing which entry was requested. In the traditional setting, it is well-known that such a protocol requires a server computation cost linear in the size of the database. Doubly efficient PIR (DEPIR) protocols circumvent this lower bound by allowing the server to perform offline preprocessing. In this very challenging topic, students are tasked to understand state-of-the-art DEPIR protocols and, for the most ambitious, to identify potential directions for improvement.


References:

  • https://eprint.iacr.org/2023/1510.pdf 
  • https://eprint.iacr.org/2022/1703.pdf

SM1 : Trade-offs for securing ML systems
Tutor: Samuel Marchal(samuel.marchal@aalto.fi)

The new AI regulation set many requirements for ML systems under the umbrella of the "trustworthy AI" concept. Many of these requirements interact with each other and can impact each other negatively. This seminar topic aims to identify the trade-offs that securing ML systems generate, i.e., how improving the security and resilience of ML systems against adversarial attacks impacts privacy, explainability, fairness, accuracy, etc. It will consist in a review of the existing studies on the topic and their outcomes, the identification of gaps in the existing literature and the finding of potential avenues for future research to be done to fill these gaps. If time allows, one of these research avenues can be explored.


References:

  • http://proceedings.mlr.press/v139/xu21b.html 
  • https://ojs.aaai.org/index.php/AAAI/article/view/26771

SM2 : Supply chain attacks against ML systems
Tutor: Samuel Marchal(samuel.marchal@aalto.fi)

Building ML systems exposes to many security threats. It is common to use many public and untrusted artifacts to build a ML systems, e.g., training libraries, datasets, pre-trained ML models, etc. All these could be compromised and com promise the ML systems that will leverage them. The goal of this seminar topic is to build a comprehensive taxonomy of supply chain attacks against ML system. A literature review will be done to find existing mitigations against the identified attacks. Based on this study, gaps in mitigation and recommendation for future research will be provided.


References:

  • https://atlas.mitre.org/


SM3 : Robustness assessment for ML systems
Tutor: Samuel Marchal(samuel.marchal@aalto.fi)

ML systems are vulnerable to adversarial ML attacks such as poisoning, backdooring, evasion, stealing, etc. It has been shown that improving the robustness against these attacks is complicated and in most cases one cannot make ML systems foolproof but only mitigate attacks to some extent. In this context it is useful to be able to evaluate how robust is a given ML system against a given attack. This seminar topic will consist in reviewing, methodologies, metrics and tools for robustness assessment of ML systems. A comparative study will be performed, identifying the strengths and weaknesses of the proposed methods along several evaluation criteria, focusing mostly on how useful they are for securing ML systems developed industry. Some hands-on experiments with robustness assessment tools can be performed. Based on this study, we will identify gaps and future research avenues for this topic.


References:

  • https://www.usenix.org/conference/usenixsecurity19/presentation/carlini
  • https://github.com/Azure/counterfit

TA : Privacy of internet users
Tutor: Tuomas Aura(tuomas.aura@aalto.fi)

This topic explores privacy-enhancing technologies and best practices available to computer network users. The topic can cover internet and wireless communication technologies and business and consumer use cases. Each student should learn about one aspect of communications privacy, survey about technical threats and solutions, and write a technical paper to explain them. While there is much high-quality historical literature on the topic, the main content of the paper should be on the state of the art and any new developments. Some possible subtopics for the work: • Beyond cookies: how web services track their users today • Analysis of commercial VPN services as privacy protection • How reliable is the Tor browser? • How to browse the web without leaving evidence • Network address randomization techniques and their effectiveness • What information does my computer leak to the access network? • DNS and user privacy • Pseudonymous participation in online discussion: best practices and risks • What we leak with git and how to control it • Technical survey of ad blockers • IMSI catchers and countermeasures The following topics are also possible for students who have relevant experience or knowledge: • The social aspect of anonymity and pseudonymity • Using social media responsibly: privacy perspectives • What does a network admin need to know about privacy regulation? • What employer access logs tell about the employees • Application and content identification for encrypted connections • How online games protect and violate player privacy • Privacy in Discord and online game chats • Ad blockers: social and business implications


References:

  • Pointers to literature will be provided depending on the chosen subtopic. Many key publications on privacy-enhancing technology and threats are published at PETS/PoPETs.

VH : Power and energy monitoring for sustainable large-scale computing
Tutor: Vesa Hirvisalo(vesa.hirvisalo@aalto.fi)

Energy consumption has become a major issue for the management of large scale computing. Such computing has been based on data centres connected by telecommunication networks, but recently there has been a trend to push the computing toward the network edge as data centres have ceased to scale efficiently. However, such management is hard as the systems are complex and they include many layers of abstraction that hide the actual resource usage from the levels on which applications operate. This has lead to various power monitoring systems that co-operate with power management systems. In addition to mere measuring of power and energy consumption, such systems should be aware of the related sustainability aspects including available energy-sources and the resulting carbon footprint. The task is to make a review on current methods on power and energy monitoring for sustainability aspects of large scale computing.


References:

  • A. Radovanović et al., "Carbon-Aware Computing for Datacenters," in IEEE Transactions on Power Systems, vol. 38, no. 2, pp. 1270-1280, March 2023, doi: 10.1109/TPWRS.2022.3173250. 
  •  Fanxin Kong and Xue Liu. 2014. A Survey on Green-Energy-Aware Power Management for Datacenters. ACM Comput. Surv. 47, 2, Article 30 (January 2015), 38 pages. https://doi.org/10.1145/2642708 
  •  Davide Zoni, Andrea Galimberti, and William Fornaciari. 2023. A Survey on Run-time Power Monitors at the Edge. ACM Comput. Surv. 55, 14s, Article 325 (December 2023), 33 pages. https://doi.org/10.1145/3593044

YV : Physics-Inspired Deep Learning for Climate Forecasting
Tutor: Yogesh Verma(yogesh.verma@aalto.fi)

Climate prediction has conventionally leaned on intricate numerical simulations rooted in atmospheric physics. While recent advancements, particularly the application of transformers in deep learning, have disrupted the simulation-centric approach by tackling challenging weather forecasts. These models often function as data-driven black boxes, overlook the climate system's intrinsic physics, and lack robust mechanisms for uncertainty quantification. This project seeks to address these limitations by exploring avenues to incorporate more physical information into climate forecasting. The initial phase involves a literature review of existing deep learning-based weather forecasting methods [1,2,3], aiming to dissect the critical components of these approaches. Subsequently, if time permits, we will evaluate these components in more realistic and dynamic real-world scenarios, aiming to enhance the accuracy and interpretability of climate predictions while preserving the richness of physical insights.


References:

  • 1. ClimODE: Climate Forecasting With Physics-informed Neural ODEs, Submitted to ICLR 2024 (https://openreview.net/forum?id=xuY33XhEGR) 
  •  2. Nguyen, Tung, et al. "ClimaX: A foundation model for weather and climate." arXiv preprint arXiv:2301.10343 (2023). 
  •  3. Kochkov, Dmitrii, et al. "Neural General Circulation Models." arXiv preprint arXiv:2311.07222 (2023)

YY : Mental Health Disclosures on Social Media Platforms
Tutor: Yunhao Yuan(yogesh.verma@aalto.fi)

In recent years, social media has evolved beyond a space for updating everyday activities and interests but also for discussing more sensitive and stigmatized topics like mental health. This project will explore how individuals use social media platforms to disclose their mental health struggles and seek support. The student will be required to conduct a literature review on the latest techniques for analyzing mental health discussions on social media, understanding the patterns in these conversations, and examining how these disclosures affect the individuals involved.


References:

  • Althoff, T., Clark, K., & Leskovec, J. (2016). Large-scale analysis of counseling conversations: An application of natural language processing to mental health. *Transactions of the Association for Computational Linguistics*, *4*, 463-476. 
  •  De Choudhury, M., Sharma, S. S., Logar, T., Eekhout, W., & Nielsen, R. C. (2017, February). Gender and cross-cultural differences in social media disclosures of mental illness. In *Proceedings of the 2017 ACM conference on computer supported cooperative work and social computing* (pp. 353-369). 
  •  Yuan, Y., Saha, K., Keller, B., Isometsä, E. T., & Aledavood, T. (2023, April). Mental Health Coping Stories on Social Media: A Causal-Inference Study of Papageno Effect. In *Proceedings of the ACM Web Conference 2023* (pp. 2677-2685).


Last modified: Thursday, 11 January 2024, 1:29 PM