Conf42 Machine Learning 2023

Schedule, Lineup & RSVP ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Upcoming CFPs ➤ https://www.papercall.io/events?cfps-scope=&keywords=conf42 0:00 Intro ➤ Sponsors & Partners mlops 0:50 Noa Goldman 1:19 Alejandro Cantos tools 2:03 Tamara Janina Fingerlin 2:53 Akmal Chaudhri 3:13 Matt Harrison models 3:39 Argo Saakyan 4:04 Logesh Kumar Umapathi 4:35 Abhiram Ravikumar & Jaspal Singh Jhass AI 5:25 Gaurab Patra 6:09 Yogesh Seenichamy deep dive 6:30 Daniel Svonava (no intro) 6:45 Zain Hasan 7:07 David Kjerrumgaard 7:42 Thank you, join Discord ➤ https://discord.gg/DnyHgrC7jC

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Noa_Goldman_hitchhiker_guide_mlops_experience Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 intro 0:22 preface 0:30 ml engineers' lives can be challenging 2:31 the current solution 4:08 who is noa 5:09 what is mlops 6:06 mlops is important to ml engineers 6:44 the main components mlops should cover 7:07 data challenges 8:22 data versioning 10:26 data visualisations 14:08 experiments challenges 15:22 experiment tracking 17:29 the main components mlops should cover 17:41 models challenges 19:13 model registry 21:09 general ux tips 23:12 what is your favourite mlops experience? tell me!

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Alejandro_Cantos_introduction_mlops_edge Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Barbara ➤ https://barbaraiot.com/ Chapters 0:00 intro 0:22 preface 0:33 the cloud is broken 1:02 ¿why is it broken?... 1:14 ...dependence on the internet 1:43 ...high latency 2:15 ...service availability 2:37 ...loss of control 3:19 ...cost management 4:14 ...vendor lock-in 5:04 ...security and privacy 5:43 ...compliance 6:25 ¿which is the solution? #edge is the new cloud 7:04 ¿what is edge computing? 8:54 edge computing drivers... 9:03 ...low latency 9:56 ...lower bandwidth consumption 10:48 ...increased privacy and security 11:22 ...greater availability and autonomy 12:13 different types of edge 15:07 impact radar for 2023 15:37 index 16:14 acciona: intelligent virtual measurement of water components 19:55 edp: energy flexibility in self-consumption systems 23:55 edge computing ♥ ai - the architecture of alexa 28:36 maybe the cloud is not broken... maybe it just needs some friends 28:56 edge ♥ cloud ♥ ai - a powerful combination 29:12 ¿how do we manage this complexity? 29:43 mlops - machine learning operations 30:04 edge ♥ cloud ♥ ai - mlops cycle 30:52 mlops challenges 32:07 architecture and frameworks... 32:17 ...edge computing platforms... 34:43 ...native providers 35:23 barbara 35:33 technological stack 40:16 demo 45:32 applications 46:20 thank you

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Tamara_Janina_Fingerlin_orchestrating_workflows_apache_airflow Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Project ➤ https://github.com/TJaniF/airflow-ml-pipeline-image-classification Chapters 0:00 intro 0:22 preface 0:31 overview 1:53 ml orchestration ∈ [ml ops] 3:21 automatable components 4:47 airflow crash course 4:55 what is apache airflow? 5:59 airflow ui 6:58 dags - tasks - operators 10:02 dags complex as you want 10:26 why airflow? 13:01 the data 14:20 sometimes it is (relatively) easy 15:37 sometimes it is harder 15:53 the pipeline 16:12 the tools 17:11 8 dags, 6 datasets 19:59 @continuous 20:39 two dags waiting for new train/test data 21:18 deferrable operators can save resources! 23:30 dynamic tasks 25:00 2 dags handling preprocessing 27:04 astro sdk - part 1 28:36 train the model 29:13 wrapping model fine-tuning into a custom operator 30:37 get a baseline 31:11 wrapping model testing into a custom operator 31:35 test fine-tuned model 32:02 airflow notifiers 33:29 customized slack alerts 34:11 deploy the best model - astro sdk part 2 36:08 demo 42:12 the results 43:17 what is next? 44:56 airflow ♥ ml - resources 48:38 thank you

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Akmal_Chaudhri_webassembly_indatabase_ml Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 introduction 0:17 agenda 4:39 install the software 7:03 initialise the source tree 7:31 create the interface Definition file 8:22 implement and compile 10:56 deploy wasm 12:25 run in the database 16:39 summary 17:14 resources

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Matt_Harrison_polars_optimized_dataframe_library Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 intro 0:22 preface 1:00 relevant background 1:38 outline of opinions 2:32 types 3:03 ints 5:41 strings 6:46 extract ffs, speed & manual 9:16 dates 10:24 chain 12:30 don't apply (if you can) 13:55 master aggregation 18:34 summary

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Argo_Saakyan_cascade_models_computer_vision_accuracy_performance Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 Intro 0:22 preface 0:48 classification 1:05 object detection task 2:44 difficulties 4:23 cascade example 6:19 why does it work? 7:38 metrics 7:55 dataset 9:24 deployment 10:45 examples 11:58 thank you!

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Logesh_Kumar_Umapathi_reasoning_planning_abilities_language_models Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Further read ➤ https://pakodas.substack.com/p/how-to-leverage-emergent-abilities?utm_source=post-email-title&publication_id=54538&post_id=117114977&isFreemail=true&utm_medium=email ➤ https://www.shaped.ai/blog/do-large-language-models-llms-reason Chapters 0:00 intro 0:22 preface 0:45 about logesh 1:16 agenda 2:14 what is reasoning? 2:56 how is reasoning measured (in the literature)? 5:32 eliciting reasoning 7:47 chain of thought prompting and self consistency 10:53 program-aided language models 13:01 plan-and-solve prompting 18:07 star: self-taught reasoner bootstrapping reasoning with reasoning 19:33 specializing smaller language models towards multi-step reasoning 20:48 distilling step-by-step 22:34 recursive and iterative prompting 23:16 least-to-most prompting 25:11 plan, eliminate, and track 27:23 describe, explain, plan and select 30:06 tool usage 30:43 react: reason and act 33:02 chameleon 37:52 acknowledgement & further reading

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Abhiram_Ravikumar_Jaspal_Singh_Jhass_data_to_discovery_clustering_bertopic_top Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Project ➤ https://github.com/abhi12ravi/BERTopic_Conf42 Chapters 0:00 intro 0:22 preface 0:36 who are we? 1:42 agenda 2:30 topic modeling use case 4:02 why bertopic? 6:47 bertopic end-to-end flow 7:36 clustering 8:33 dataset description 8:56 demo 13:07 what is hdbscan? 13:37 to understand hdbscan we need to know dbscan 15:39 what if there was no fixed radius? 15:54 k-nn algorithm to define radius 18:00 minimum spanning tree finds density and hierachy 19:49 density based spatial clustering 20:09 stability score "λ" 22:03 final clusters 22:32 hdbscan steps 23:05 hdbscan - performance comparison 23:58 hdbscan - strenghts and weaknesses 24:47 conclusion and future scope 26:28 references & ressources 26:34 thank you

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Gaurab_Patra_generative_ai_crowdsourced_interactivity Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 Intro 0:22 preface 1:05 what does that mean? 1:22 human-to-human communication 2:03 human-to-machine communication 6:25 generative ai... 8:29 ...generative adversarial networks (gans) 10:45 ...language generative models 14:38 gan architecture 15:59 process and understand human language at scale 16:29 gans enables generating input-like output 17:51 generative ai is at the foundation building stage 19:41 crowdsourcing for interactivity 22:45 image synthesis 23:48 image extension 24:33 unicorner - startup idea validator 28:32 coperate with machines, don't operate

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Yogesh_Seenichamy_ml_healthcare Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Heart Failure Predition Dataset ➤ https://www.kaggle.com/datasets/fedesoriano/heart-failure-prediction Code ➤ https://colab.research.google.com/drive/1Jq9pszNpVk8qW9CrwGwapHB2j4xipgcH?usp=sharing Chapters 0:00 intro 0:22 preface 0:33 topics 1:10 heart failure background 2:38 current limitations of diagnosis 3:09 neural networs 5:14 heart failure prediction dataset (kaggle) 6:29 demo 15:25 conclusion 15:47 thank you

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Daniel_Svonava_vector_ops_embeddingpowered_apps Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 intro 0:22 preface 0:36 building vector-powered apps 1:35 what did we lose with language? 4:14 natural language is a bottleneck 4:50 natural language is ambiguous 5:56 vectors are better! (mostly) 7:55 search before vectors 10:32 recommendations before vectors 14:42 search & recommendations with vectors 17:40 approximate nearest neighbours? 20:23 it can be measured 22:57 building the content vectors... 25:39...and user vectors in the same space! 27:53 let ann do the heavy lifting 29:39 query manager on top 32:36 what will you need to get started & MVP? 38:26 towards a #vectorops platform 42:57 what about generative ai?! 43:41 chat and chains! 45:30 what the frig? 46:42 agents & memory! 48:41 let's connect!

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_Zain_Hasan_vector_databases Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 intro 0:22 preface 1:00 about zain 1:42 from keyword search to semantic search 4:31 process, understand and search through unstructured data... 5:01 ...in a scalable and secure way 5:20 use ml to understand the context 5:57 machine learning models 9:14 vector representations 10:54 vector databases 17:16 weaviate: an open-source vector db 18:01 a vector search pipeline 19:40 weaviate is modular: flexible search pipelines 21:17 using vector search to provide llms context 25:12 demo 43:49 reach out

Read the abstract ➤ https://www.conf42.com/Machine_Learning_2023_David_Kjerrumgaard_event_streaming_apps_java_microservices Other sessions at this event ➤ https://www.conf42.com/ml2023 Join Discord ➤ https://discord.gg/DnyHgrC7jC Chapters 0:00 intro 0:22 preface 0:38 about the speaker 1:06 agenda 1:27 event-driven microservices 3:00 what is apache pulsar? 3:42 pulsar pub/sub model 4:15 topics 5:20 physical architecture of a pulsar cluster 7:16 what are pulsar functions? 8:15 pulsar functions programming model 9:23 why pulsar functions? 10:32 when to use pulsar functions 11:21 developing pulsar functions - sdk 12:32 packaging functions 13:14 coding demo 25:24 review 26:06 let's keep in touch