August 2022 - Challenge2022

[Challenge2022] Specific session for challenge participants (GNNet co-located with ACM CoNEXT 2022)

by Pere Barlet

Dear challenge participants, We are very excited to announce that this year we are organizing the first workshop on GNN applied to networks (GNNet 2022) co-located with ACM CoNEXT 2022 (Core Rank A): https://bnn.upc.edu/workshops/gnnet2022/ We encourage all challenge participants (either from this edition or past editions) to submit your challenge solution, or any other GNN-related research, to the GNNet workshop. All accepted papers will be included in the conference proceedings and be made available in the ACM Digital Library. We plan to organize a *specific session for challenge participants* in the workshop. You can find below the general CFP. Please note deadlines are very tight (we have to strictly adhere to CoNEXT timings) so, in order to be considered for publication, your paper should be submitted on*September 16* (before the current edition of the challege ends). Also note that your solution does not necessarily be in the top of the ranking for the paper to be accepted for publication. We hope this workshop can serve to build a small community among those of us interested in what GNN can bring to networking! Do not hesitate to contact us if you have any questions or doubts! Pere. (on behalf of all workshop chairs and challenge organizers) -- CALL FOR PAPERS 1st Graph Neural Networking Workshop (GNNet) Co-located with ACM CoNEXT 2022 Rome, Italy, December 9, 2022 https://bnn.upc.edu/workshops/gnnet2022 We are glad to announce the first edition of the “Graph Neural Networking Workshop 2022”, which is organized as part of ACM CoNEXT 2022, to be held in Rome, Italy. All accepted papers will be included in the conference proceedings and be made available in the ACM Digital Library. IMPORTANT DATES =============== Paper registration deadline: September 9, 2022 Paper submission deadline: September 16, 2022 Paper acceptance notifications: October 17, 2022 Camera ready due: October 25, 2022 MOTIVATION ========== While AI/ML is today mainstream in domains such as computer vision and speech recognition, traditional AI/ML approaches have produced below-par results in many networking applications. Proposed AI/ML solutions in networking do not properly generalize, can be unreliable and ineffective in real-network deployments, and are in general unable to properly deal with the strong dynamics and changes (i.e., concept drift) occurring in networking applications. Graphs are emerging as an abstraction to represent complex data. Computer Networks are fundamentally graphs, and many of their relevant characteristics – such as topology and routing – are represented as graph-structured data. Machine learning, especially deep representation learning, on graphs is an emerging field with a wide array of applications. Within this field, Graph Neural Networks (GNNs) have been recently proposed to model and learn over graph-structured data. Due to their unique ability to generalize over graph data, GNNs are a central tool to apply AI/ML techniques to networking applications. GOALS ===== The goal of GNNet is to leverage graph data representations and modern GNN technology to advance the application of AI/ML in networking. GNNet provides the first dedicated venue to present and discuss the latest advancements on GNNs and general AI/ML on graphs applied to networking problems. GNNet will bring together leaders from academia and industry to showcase recent methodological advances of GNNs and their application to networking problems, covering a wide range of applications and practical challenges for large-scale training and deployment. We expect GNNet would serve as the meeting point for the growing community on this fascinating domain, which has currently not a specific forum for sharing and discussion. The GNNet workshop seeks for contributions in the field of GNNs and graph-based learning applied to data communication networking problems, including the analysis of on-line and off-line massive datasets, network traffic traces, topological data, cybersecurity, performance measurements, and more. GNNet also encourages novel and out-of-the-box approaches and use cases related to the application of GNNs in networking. The workshop will allow researchers and practitioners to discuss the open issues related to the application of GNNs and graph-based learning to networking problems and to share new ideas and techniques for big data analysis and AI/ML in data communication networks. TOPICS OF INTEREST ================== We encourage both mature and positioning submissions describing systems, platforms, algorithms and applications addressing all facets of the application of GNNs and Machine learning on graphs to the analysis of data communication networks. We are particularly interesting in disruptive and novel ideas that permit to unleash the power of GNNs in the networking domain. The following is a non-exhaustive list of topics: - GNNs and graph-based learning in networking applications - Representation Learning on networking-related graphs - Application of GNNs to network and service management - Application of GNNs to network security and anomaly detection - Application of GNNs to detection of malware, botnets, intrusions, phishing, and abuse detection - Adversarial learning for GNN-driven networking applications - GNNs for data generation and digital twining in networking - Temporal and dynamic GNNs in networking - Graph-based analytics for visualization of complex networking applications - Libraries, benchmarks, and datasets for GNN-based networking applications - Scalability of GNNs for networking applications - Explainability, fairness, accountability, transparency, and privacy issues in GNN-based networking - Challenges, pitfalls, and negative results in applying GNNs to networking applications SPECIAL SESSION =============== GNNet would include a dedicated special session where the top teams competing at the third edition of the Graph Neural Networking Challenge (https://bnn.upc.edu/challenge/gnnet2022/) would be invited to present the winning solutions of the challenge, providing an excellent complement to the main program. SUBMISSION INSTRUCTIONS ======================= Submissions must be original, unpublished work, and not under consideration at another conference or journal. Submitted papers must be at most six (6) pages long, including all figures, tables, references, and appendices in two-column 10pt ACM format. Papers must include authors names and affiliations for single-blind peer reviewing by the PC. Authors of accepted papers are expected to present their papers at the workshop. All accepted papers will be included in the conference proceedings and be made available in the ACM Digital Library. WORKSHOP CHAIRS ================ Pere Barlet-Ros, BNN-UPC, Spain Pedro Casas, AIT, Austria Franco Scarselli, University of Siena, Italy Xiangle Cheng, Huawei, China Albert Cabellos, BNN-UPC, Spain PRELIMINARY PC COMMITTEE ======================== Lilian Berton, University of Sao Paulo, Brazil Albert Bifet, Télécom ParisTech & University of Waikato, New Zealand Laurent Ciavaglia, Rakuten, Japan Constantine Dovrolis, Georgia Tech, USA Lluís Fàbrega, UdG, Spain Jerome François, INRIA, France Fabien Geyer, Technical University of Munich, Germany Matthias Herlich, Salzburg Research, Austria Zied Ben Houidi, Huawei Technologies, France Wolfgang Kellerer, Technical University of Munich, Germany Federico Larroca, Universidad de la República, Uruguay Alina Lazar, Youngstown State University, USA Gonzalo Mateos, University of Rochester, USA Diego Perino, Telefonica Research, Spain Alejandro Ribeiro, University of Pennsylvania, USA Krzysztof Rusek, AGH University of Science and Technology, Poland José Suárez-Varela, BNN-UPC, Spain Stefano Traverso, Ermes Cyber Security, Italy

2 years, 5 months

1
3
0 0

[Challenge2022] data generation time issue

by limeiyjy＠chinamobile.com

Hi， Q1: The constraints for the training dataset show "Average bandwidth must be between 10 and 10000". I changed the "max_avg_lbda = 1000" parameter in quickstart notebook to "max_avg_lbda = 10000". The time taken to generate data has increased from about 10 minutes to 4 hours. I wonder if there is any problem. Q2：Whether the submitted model for testing can be the best model generated in the training process. Or must be a model after 20 epochs of training. Could you please advise? Thanks， Mei

2 years, 7 months

2
1
0 0

[Challenge2022] error training with WFQ

by yawi yoga

Hi, After changing scheduling policy to WFQ in the training data, the training script gives the following error when trying to train a model: Node: 'gradients/rnn_13/TensorArrayUnstack/TensorListFromTensor_grad/TensorListStack' Operation expected a list with 1 elements but got a list with 2 elements. [[{{node gradients/rnn_13/TensorArrayUnstack/TensorListFromTensor_grad/TensorListStack}}]] [Op:__inference_train_function_32315] I created the data using quickstart notebook, the only change is replacing this line in generate_topology(): G.nodes[i]["schedulingPolicy"] = "FIFO" with this G.nodes[i]["schedulingPolicy"] = "WFQ" G.nodes[i]["schedulingWeights"] = '25,25,50' The docker runs OK on this data but I can't train a model on it due to the above error. If the policies are FIFO, the training runs without issues. Could you please let me know how to resolve this? Thanks, Yawi ----- here is the full traceback for your reference: INFO: Starting training from scratch... Epoch 1/20 Traceback (most recent call last): File "/home/yawi/dev/GNNetworkingChallenge/api_test.py", line 23, in <module> main(data_dir, final_evaluation=True) File "/home/yawi/dev/GNNetworkingChallenge/RouteNet_Fermi/__init__.py", line 101, in main model.fit(ds_train, File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/utils/traceback_utils.py", line 67, in error_handler raise e.with_traceback(filtered_tb) from None File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/tensorflow/python/eager/execute.py", line 54, in quick_execute tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name, tensorflow.python.framework.errors_impl.InvalidArgumentError: Graph execution error: Detected at node 'gradients/rnn_13/TensorArrayUnstack/TensorListFromTensor_grad/TensorListStack' defined at (most recent call last): File "/home/yawi/dev/GNNetworkingChallenge/api_test.py", line 23, in <module> main(data_dir, final_evaluation=True) File "/home/yawi/dev/GNNetworkingChallenge/RouteNet_Fermi/__init__.py", line 101, in main model.fit(ds_train, File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/utils/traceback_utils.py", line 64, in error_handler return fn(*args, **kwargs) File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/engine/training.py", line 1409, in fit tmp_logs = self.train_function(iterator) File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/engine/training.py", line 1051, in train_function return step_function(self, iterator) File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/engine/training.py", line 1040, in step_function outputs = model.distribute_strategy.run(run_step, args=(data,)) File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/engine/training.py", line 1030, in run_step outputs = model.train_step(data) File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/engine/training.py", line 893, in train_step self.optimizer.minimize(loss, self.trainable_variables, tape=tape) File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/optimizers/optimizer_v2/optimizer_v2.py", line 537, in minimize grads_and_vars = self._compute_gradients( File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/optimizers/optimizer_v2/optimizer_v2.py", line 590, in _compute_gradients grads_and_vars = self._get_gradients(tape, loss, var_list, grad_loss) File "/home/yawi/.conda/envs/gnnch/lib/python3.9/site-packages/keras/optimizers/optimizer_v2/optimizer_v2.py", line 471, in _get_gradients grads = tape.gradient(loss, var_list, grad_loss) Node: 'gradients/rnn_13/TensorArrayUnstack/TensorListFromTensor_grad/TensorListStack' Operation expected a list with 1 elements but got a list with 2 elements. [[{{node gradients/rnn_13/TensorArrayUnstack/TensorListFromTensor_grad/TensorListStack}}]] [Op:__inference_train_function_32315] Process finished with exit code 1

2 years, 7 months

2
1
0 0

[Challenge2022] Reminder: Round table for the Graph Neural Networking Challenge [Thu. Aug 4, at 14:00 – 15:00 CEST]

by José Suárez-Varela

Dear participants, Today (Aug. 4), there will be a round table for the Graph Neural Networking Challenge 2022. I will be happy to answer questions, receive feedback or give some tips. All teams participating in the challenge are encouraged to attend this meeting. *Date*: Thu., August 4, 2022 *Time*: 14:00 – 15:00 CEST *Meeting ID (Zoom)*: 997 9878 1690 *Registration*: https://itu.zoom.us/meeting/register/tJ0qfuGvpjosE9RdrWxt0rC0r8h1PTJAjn9F *Please, note you need to register in advance to join the meeting.* *Description*: This round table is scheduled for participants of the Graph Neural Networking Challenge 2022: "Improving Network Digital Twins through Data-centric AI" to have an open platform with the organizers and host to ask questions, share their experience and any other discussion related to the problem statement. *Agenda*: 1. Brief introduction of the problem statement. 2. Resources and materials available for participants. 3. Brief overview of Jupyter Notebooks. 4. Q&A and interaction with participants or Teams. Best regards, José Suárez-Varela ------------------------ Postdoctoral Researcher Barcelona Neural Networking center (BNN-UPC) Universitat Politècnica de Catalunya

2 years, 7 months

1
0
0 0

[Challenge2022] error probability don't sum to 1

by yawi yoga

Hi, When running the docker it often complains that packet size probabilities don't sum to 1. For example: "ERROR:root:0: Error using traffic matrix file /data/tm/tm_00000_00000.txt at line 22. The sum of probabilities of all packets sizes should be one" In the above example, the line it complains about is the following: 3,3,5610,0,0,760,0.44,1878,0.48,595,0.08,1 After some checks it seems that the error is due to exact comparison of float sum of probabilities to 1, which may be problematic with floats - e.g. 0.44 + 0.48 + 0.08 is 0.99999999999 and so statement "(0.44 + 0.48 + 0.08) == 1" returns False Is there a newer version of the docker with this fixed?

2 years, 7 months

2
1
0 0

[Challenge2022] Round table for the Graph Neural Networking Challenge [Thu. Aug 4, at 14:00 – 15:00 CEST]

by José Suárez-Varela

Dear participants, Next Thursday (Aug. 4) there will be a round table for the Graph Neural Networking Challenge 2022. I will be happy to answer questions, receive feedback or give some tips. I hope it can be helpful for all of you. Please, save the date! *Date*: Thu., August 4, 2022 *Time*: 14:00 – 15:00 CEST *Meeting ID (Zoom)*: 997 9878 1690 *Registration*: https://itu.zoom.us/meeting/register/tJ0qfuGvpjosE9RdrWxt0rC0r8h1PTJAjn9F *Please, note you need to register in advance to join the meeting.* *Description*: This round table is scheduled for participants of the Graph Neural Networking Challenge 2022: "Improving Network Digital Twins through Data-centric AI" to have an open platform with the organizers and host to ask questions, share their experience and any other discussion related to the problem statement. *Agenda*: 1. Brief introduction of the problem statement. 2. Resources and materials available for participants. 3. Brief overview of Jupyter Notebooks. 4. Q&A and interaction with participants or Teams. José Suárez-Varela ------------------------ Postdoctoral Researcher Barcelona Neural Networking center (BNN-UPC) Universitat Politècnica de Catalunya

2 years, 7 months

1
0
0 0

2025

2024

2023

2022

Challenge2022 August 2022