[HN Gopher] A Critical Field Guide for Working with Machine Lear...
       ___________________________________________________________________
        
       A Critical Field Guide for Working with Machine Learning Datasets
        
       Author : Amorymeltzer
       Score  : 76 points
       Date   : 2023-02-16 12:58 UTC (1 days ago)
        
 (HTM) web link (knowingmachines.org)
 (TXT) w3m dump (knowingmachines.org)
        
       | clircle wrote:
       | 100 years ago statisticians starting working on the theories of
       | survey design and experimental design to clarify what types, how
       | much, what mixture, and how to collect data. Machine learning
       | people could learn a lot from these old, uncool theories if they
       | want to approach data with rigor. That one can approach data
       | collection and usage in a systematic way is totally lost now,
       | it's just web scraping.
        
         | kordlessagain wrote:
         | A system for this would allow individual collections of
         | documents with discussions attached. Without people, the
         | machines won't know where to put attention. That may change
         | when they have individual identity, but for now the work
         | probably needs to be done cooperatively.
        
         | steppi wrote:
         | This hasn't been my experience. I don't think this knowledge
         | has been lost and it's often still applied when incentivized,
         | such as when quality output of machine learning models serves
         | an immediate business need, or in scientific applications of
         | machine learning. By your reference to web scraping I assume
         | your referring to datasets for large language models, but I
         | don't think the kitchen sink approach is applied out of
         | ignorance, but out of a principled decision to forgo data
         | hygiene in order to collect datasets large enough to feed such
         | large models. Also, strictly speaking, one only needs to apply
         | the critical approach when constructing and using validation
         | datasets, and it's typical for larger noisier training datasets
         | to outperform smaller more rigorous ones when models trained on
         | both are tested on carefully constructed validation datasets.
        
         | razor_router wrote:
         | I definitely agree that taking a systematic approach to data
         | collection and usage is incredibly important. However, I think
         | it is impossible to simply dismiss the ways that modern machine
         | learning has revolutionized the way we understand and interact
         | with data. There is no question that rigorous survey design and
         | experimental design theories still have merit and should be
         | taken into account, but I think machine learning has allowed us
         | to delve deeper into data analysis and uncover patterns and
         | insights that would not be possible without its advancements.
        
         | KRAKRISMOTT wrote:
         | Machine learning (or more specifically _deep learning_ ) is
         | shaking the world precisely because the models at the highest
         | end are able to take in mostly unlabeled garbage and spit out
         | useful results in an unsupervised manner. The world is not full
         | of i.i.d. spherical cows.
         | 
         | 100 years of statistical advances have not been able to solve
         | image recognition at scale until convolutional neural networks.
         | Large language models may be _somewhat_ Bayesian but they do
         | not resemble any traditional Bayesian techniques except at the
         | highest level. Clean data and good sampling are incompatible
         | with the real world outside of laboratory conditions (or large
         | budgets).
        
           | avs733 wrote:
           | That is a fairly brave statement given some of the research
           | looking at how failed attempts at ML modeling fall apart [1]
           | 
           | [1] https://ai.googleblog.com/2021/06/data-cascades-in-
           | machine-l...
        
           | wongarsu wrote:
           | Some of them do, but others depend on labeled and augmented
           | data. For example training facial recognition starts with
           | images labeled by who appears in them, then you run a model
           | to find the face in the image and crop to that, then another
           | image tells you where the eyes are and you use that to rotate
           | the image to make the eyes level, and that's the input to the
           | actual facial recognition neural network. And that's for the
           | "naive" approaches that don't have special handling for faces
           | don't look at the camera. And the field still routinely has
           | problems because it turns out that somebody's data set only
           | contained certain lighting conditions, or severely
           | underrepresented certain races, etc.
        
             | KRAKRISMOTT wrote:
             | Not sure about image recognition but for the language
             | models and audio space, semi/self-supervised methods are
             | currently doing quite well.
        
         | dachryn wrote:
         | you are too pessimistic. Senior profiles definitely dived into
         | the statistics aspect of it all. Experiment design is often a
         | big part of modern business cases, because naturally it is part
         | of the cost to obtain data.
         | 
         | It is still taught at university if you do a master degree or
         | PhD at a decent place.
         | 
         | But I agree that many juniors have no clue. Especially those
         | who are self taught and join in from a different field
        
       | m00dy wrote:
       | terrible layout to read..
        
       | peanutcrisis wrote:
       | Note that the term "Critical" here refers to the likes of
       | "Critical Theory", "Critical Social Justice", "Critical Race
       | Theory" etc. It's biased from the offset so I'd take anything
       | from this site with a pinch of salt.
        
       | jna_sh wrote:
       | I love the faux spreadsheet presentation of this guide.
        
         | firatsarlar wrote:
         | +1
        
         | r3trohack3r wrote:
         | I didn't realize it wasn't a "real" spreadsheet application
         | until your comment.
         | 
         | The design gave me a bit of anxiety, it was so convincing I was
         | worried about clicking into one of the cells!
         | 
         | Anxiety included - I also love the layout.
        
       | kordlessagain wrote:
       | Interesting it doesn't cover technologies for storing data. It
       | does mention analytical operations though.
       | 
       | Anyone have a suggestion for storing large amounts of ML data,
       | training sets or otherwise? I've been using FeatureBase and
       | weaviate, and would be interested in learning about other
       | solutions.
        
         | kacperlukawski wrote:
         | If you want to perform vector search over your data, then
         | Qdrant (https://qdrant.tech) is worth checking out.
        
         | gschoeni wrote:
         | We've been working on an open source tool called Oxen to help
         | store large ML datasets. It's optimized for large sets of
         | unstructured data ie images, video, audio, text, as well as
         | parquet or arrow style DataFrames.
         | 
         | Would love to get some feedback on it!
         | 
         | https://github.com/Oxen-AI/oxen-release#-oxen
        
         | wongarsu wrote:
         | "ML data" could be just about anything: images, sounds, graphs,
         | tabular data, vectors. Depending on what your data is, you
         | might end up with very different storage solutions.
         | 
         | Though I think one common approach is to just dump most data in
         | an S3-compatible datastore of your choice (there's seaweedfs or
         | ceph, or the cloud provider of your choice). Specialized
         | databases make a lot of sense when you need features like
         | vector similarity search though.
        
         | fzliu wrote:
         | Vector databases in general are good for storing large amounts
         | of unstructured data by first converting them into embeddings
         | via ML models. There's also feature stores, which store and
         | organize features for later use in model training or predictive
         | analytics. Feature stores generally come in _before_ models get
         | trained, while vector databases generally come _after_ (i.e.
         | they use trained models).
         | 
         | Milvus (https://milvus.io) and Feast (https://feast.dev/) are
         | two of the most well known vector databases and feature stores,
         | respectively.
        
       ___________________________________________________________________
       (page generated 2023-02-17 23:02 UTC)