JSR 381 (Visual Recognition in Java) and Apache Ignite

[Vladimir Ozerov-2]

Hi Alexey, Igniters,

Let me introduce you Heather, Zoran, and Frank. Heather is the Chair of the
JCP. Zoran and Frank are JSR 381 spec leads. They are interested in
discussing the upcoming visual recognition specification with the Apache
Ignite community, to understand whether the community has any interest in
implementing it.

Zoran and Frank,

Please meet Alexey Zinoviev, who is the principal maintainer of the Apache
ML module. I hope he will be able to help you with your questions.

Regards,
Vladimir.

[Alexey Zinoviev]

Nice to see all of you guys (thanks Vladimir) here on dev-list of Apache
Ignite that includes distributed ML.

First of all, want to say thank you for your great work to build specs to
the Java world, hope it will be helpful for Java devs in the nearest
future.
In the second, a few days ago I found a link
<https://twitter.com/frankgreco/status/1242885726196424712> in Frank Greco
twitter to this JSR and read a few examples to understand the proposed API.
Also, we (in Apache Ignite ML) have experimental integration with
TensorFlow for inference and distributed training purposes and I'm
interested in TF Java API development  and DL4j/PyTorch possible
integration.
Will be great to have common API for different implementations.

P.S. From my side, I've ready to join your team if the specification is not
finished yet, because I have a few comments and thoughts based on my
experience of different API usage. A few weeks ago I also joined to the
developer group of JVM API for TensorFlow (SIG JVM), hope it will be
helpful in both directions to communicate each other.



пн, 30 мар. 2020 г. в 21:00, Vladimir Ozerov <[hidden email]>:

> Hi Alexey, Igniters,
>
> Let me introduce you Heather, Zoran, and Frank. Heather is the Chair of
> the JCP. Zoran and Frank are JSR 381 spec leads. They are interested in
> discussing the upcoming visual recognition specification with the Apache
> Ignite community, to understand whether the community has any interest in
> implementing it.
>
> Zoran and Frank,
>
> Please meet Alexey Zinoviev, who is the principal maintainer of the Apache
> ML module. I hope he will be able to help you with your questions.
>
> Regards,
> Vladimir.
>

[Zoran Sevarac]

On Tue, Mar 31, 2020 at 3:03 PM Alexey Zinoviev <[hidden email]>
wrote:

> Nice to see all of you guys (thanks Vladimir) here on dev-list of Apache
> Ignite that includes distributed ML.
>

Nice to meet you too Alexey, and thanks to Vladimir for introducing us to
this community.

> First of all, want to say thank you for your great work to build specs to
> the Java world, hope it will be helpful for Java devs in the nearest
> future.
> In the second, a few days ago I found a link
> <https://twitter.com/frankgreco/status/1242885726196424712> in Frank
> Greco twitter to this JSR and read a few examples to understand the
> proposed API.
> Also, we (in Apache Ignite ML) have experimental integration with
> TensorFlow for inference and distributed training purposes and I'm
> interested in TF Java API development  and DL4j/PyTorch possible
> integration.
> Will be great to have common API for different implementations.
>

Great, we're interested to hear about your experience with these libs, and
requirements for distributed training. I also saw
https://apacheignite.readme.io/docs/machine-learning  it looks like
something different from Tensorflow integration? I guess it would make
sense for end-users to use the same API for the same task regardless of the
backend library.
By the way, I recommend that you take a look at https://djl.ai it's deep
learning library from Amazon which is a Java wrapper around MxNet, TF and
PyTorch, they also participate in JSR381, and have provided initial
implementation.


> P.S. From my side, I've ready to join your team if the specification is
> not finished yet, because I have a few comments and thoughts based on my
> experience of different API usage. A few weeks ago I also joined to the
> developer group of JVM API for TensorFlow (SIG JVM), hope it will be
> helpful in both directions to communicate each other.
>

Sounds good,  at this phase we're looking for the input and feedback from
various communities in order to make API that will meet their needs.
So we would definitely like to hear your thoughts and we're open to adapt
the specification.

Best regards
Zoran

>
> пн, 30 мар. 2020 г. в 21:00, Vladimir Ozerov <[hidden email]>:
>
>> Hi Alexey, Igniters,
>>
>> Let me introduce you Heather, Zoran, and Frank. Heather is the Chair of
>> the JCP. Zoran and Frank are JSR 381 spec leads. They are interested in
>> discussing the upcoming visual recognition specification with the Apache
>> Ignite community, to understand whether the community has any interest in
>> implementing it.
>>
>> Zoran and Frank,
>>
>> Please meet Alexey Zinoviev, who is the principal maintainer of the
>> Apache ML module. I hope he will be able to help you with your questions.
>>
>> Regards,
>> Vladimir.
>>
>

--
Zoran Sevarac, PhD, Associate Professor
University of Belgrade, Faculty of Organisational Sciences, Department for
Software Engineering
Java Champion <https://community.oracle.com/community/java/java-champions>
| Oracle Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
Co-founder & CEO

[Alexey Zinoviev]

We removed the TF integration from the current release due to a few
problems with TF 2.0 and new policy about tf.contrib but it will turn out
after problem solving
You could find docs for previous Ignite and TF versions
https://apacheignite.readme.io/v2.7.6/docs/tensor-flow

I look on the https://djl.ai, it seems a little bit different with JSR (but
close in builders), also as I understand (looking in sources) they don't
support TF engine yet
Is it a mature framework, how do you think?

Happy to join the unification movement in Java DL, where is the best place
to leave feedback about specification? Here, in this thread or on another
mailing list?

Sincerely yours,
              Alexey Zinoviev

ср, 1 апр. 2020 г. в 16:59, Zoran Sevarac <[hidden email]>:

> On Tue, Mar 31, 2020 at 3:03 PM Alexey Zinoviev <[hidden email]>
> wrote:
>
>> Nice to see all of you guys (thanks Vladimir) here on dev-list of Apache
>> Ignite that includes distributed ML.
>>
>
> Nice to meet you too Alexey, and thanks to Vladimir for introducing us to
> this community.
>
>
>> First of all, want to say thank you for your great work to build specs to
>> the Java world, hope it will be helpful for Java devs in the nearest
>> future.
>> In the second, a few days ago I found a link
>> <https://twitter.com/frankgreco/status/1242885726196424712> in Frank
>> Greco twitter to this JSR and read a few examples to understand the
>> proposed API.
>> Also, we (in Apache Ignite ML) have experimental integration with
>> TensorFlow for inference and distributed training purposes and I'm
>> interested in TF Java API development  and DL4j/PyTorch possible
>> integration.
>> Will be great to have common API for different implementations.
>>
>
> Great, we're interested to hear about your experience with these libs, and
> requirements for distributed training. I also saw
> https://apacheignite.readme.io/docs/machine-learning  it looks like
> something different from Tensorflow integration? I guess it would make
> sense for end-users to use the same API for the same task regardless of the
> backend library.
> By the way, I recommend that you take a look at https://djl.ai it's deep
> learning library from Amazon which is a Java wrapper around MxNet, TF and
> PyTorch, they also participate in JSR381, and have provided initial
> implementation.
>
>
>> P.S. From my side, I've ready to join your team if the specification is
>> not finished yet, because I have a few comments and thoughts based on my
>> experience of different API usage. A few weeks ago I also joined to the
>> developer group of JVM API for TensorFlow (SIG JVM), hope it will be
>> helpful in both directions to communicate each other.
>>
>
> Sounds good,  at this phase we're looking for the input and feedback from
> various communities in order to make API that will meet their needs.
> So we would definitely like to hear your thoughts and we're open to adapt
> the specification.
>
> Best regards
> Zoran
>
>
>
>>
>> пн, 30 мар. 2020 г. в 21:00, Vladimir Ozerov <[hidden email]>:
>>
>>> Hi Alexey, Igniters,
>>>
>>> Let me introduce you Heather, Zoran, and Frank. Heather is the Chair of
>>> the JCP. Zoran and Frank are JSR 381 spec leads. They are interested in
>>> discussing the upcoming visual recognition specification with the Apache
>>> Ignite community, to understand whether the community has any interest in
>>> implementing it.
>>>
>>> Zoran and Frank,
>>>
>>> Please meet Alexey Zinoviev, who is the principal maintainer of the
>>> Apache ML module. I hope he will be able to help you with your questions.
>>>
>>> Regards,
>>> Vladimir.
>>>
>>
>
> --
> Zoran Sevarac, PhD, Associate Professor
> University of Belgrade, Faculty of Organisational Sciences, Department
> for Software Engineering
> Java Champion <https://community.oracle.com/community/java/java-champions>
> | Oracle Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
> Co-founder & CEO
>

[Zoran Sevarac]

>
> I look on the https://djl.ai, it seems a little bit different with JSR
> (but close in builders), also as I understand (looking in sources) they
> don't support TF engine yet
> Is it a mature framework, how do you think?
>

It is relatively new but very promising in my opinion. TF support is not
publically available but they do have it.
And also they provide implementation of JSR381 at
https://github.com/JavaVisRec/visrec-djl
They have recently joined JSR381


> Happy to join the unification movement in Java DL, where is the best place
> to leave feedback about specification? Here, in this thread or on another
> mailing list?
>

For the initial feedback please use the official JSR381 mailing list at
https://groups.io/g/visrec

Best
Zoran

>
> Sincerely yours,
>               Alexey Zinoviev
>
> ср, 1 апр. 2020 г. в 16:59, Zoran Sevarac <[hidden email]>:
>
>> On Tue, Mar 31, 2020 at 3:03 PM Alexey Zinoviev <[hidden email]>
>> wrote:
>>
>>> Nice to see all of you guys (thanks Vladimir) here on dev-list of Apache
>>> Ignite that includes distributed ML.
>>>
>>
>> Nice to meet you too Alexey, and thanks to Vladimir for introducing us to
>> this community.
>>
>>
>>> First of all, want to say thank you for your great work to build specs
>>> to the Java world, hope it will be helpful for Java devs in the nearest
>>> future.
>>> In the second, a few days ago I found a link
>>> <https://twitter.com/frankgreco/status/1242885726196424712> in Frank
>>> Greco twitter to this JSR and read a few examples to understand the
>>> proposed API.
>>> Also, we (in Apache Ignite ML) have experimental integration with
>>> TensorFlow for inference and distributed training purposes and I'm
>>> interested in TF Java API development  and DL4j/PyTorch possible
>>> integration.
>>> Will be great to have common API for different implementations.
>>>
>>
>> Great, we're interested to hear about your experience with these libs,
>> and requirements for distributed training. I also saw
>> https://apacheignite.readme.io/docs/machine-learning  it looks like
>> something different from Tensorflow integration? I guess it would make
>> sense for end-users to use the same API for the same task regardless of the
>> backend library.
>> By the way, I recommend that you take a look at https://djl.ai it's deep
>> learning library from Amazon which is a Java wrapper around MxNet, TF and
>> PyTorch, they also participate in JSR381, and have provided initial
>> implementation.
>>
>>
>>> P.S. From my side, I've ready to join your team if the specification is
>>> not finished yet, because I have a few comments and thoughts based on my
>>> experience of different API usage. A few weeks ago I also joined to the
>>> developer group of JVM API for TensorFlow (SIG JVM), hope it will be
>>> helpful in both directions to communicate each other.
>>>
>>
>> Sounds good,  at this phase we're looking for the input and feedback from
>> various communities in order to make API that will meet their needs.
>> So we would definitely like to hear your thoughts and we're open to adapt
>> the specification.
>>
>> Best regards
>> Zoran
>>
>>
>>
>>>
>>> пн, 30 мар. 2020 г. в 21:00, Vladimir Ozerov <[hidden email]>:
>>>
>>>> Hi Alexey, Igniters,
>>>>
>>>> Let me introduce you Heather, Zoran, and Frank. Heather is the Chair of
>>>> the JCP. Zoran and Frank are JSR 381 spec leads. They are interested in
>>>> discussing the upcoming visual recognition specification with the Apache
>>>> Ignite community, to understand whether the community has any interest in
>>>> implementing it.
>>>>
>>>> Zoran and Frank,
>>>>
>>>> Please meet Alexey Zinoviev, who is the principal maintainer of the
>>>> Apache ML module. I hope he will be able to help you with your questions.
>>>>
>>>> Regards,
>>>> Vladimir.
>>>>
>>>
>>
>> --
>> Zoran Sevarac, PhD, Associate Professor
>> University of Belgrade, Faculty of Organisational Sciences, Department
>> for Software Engineering
>> Java Champion
>> <https://community.oracle.com/community/java/java-champions> | Oracle
>> Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
>> Co-founder & CEO
>>
>

--
Zoran Sevarac, PhD, Associate Professor
University of Belgrade, Faculty of Organisational Sciences, Department for
Software Engineering
Java Champion <https://community.oracle.com/community/java/java-champions>
| Oracle Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
Co-founder & CEO

[Alexey Zinoviev]

Thanks, I will check it out. Great dlj library with the reference JSR
implementation, I have not seen yet this repo.

Will return in a few weeks with suggestions on JSR mailing list

Sincerely yours,
   Alexey Zinoviev

ср, 1 апр. 2020 г., 18:49 Zoran Sevarac <[hidden email]>:

> I look on the https://djl.ai, it seems a little bit different with JSR
>> (but close in builders), also as I understand (looking in sources) they
>> don't support TF engine yet
>> Is it a mature framework, how do you think?
>>
>
> It is relatively new but very promising in my opinion. TF support is not
> publically available but they do have it.
> And also they provide implementation of JSR381 at
> https://github.com/JavaVisRec/visrec-djl
> They have recently joined JSR381
>
>
>> Happy to join the unification movement in Java DL, where is the best
>> place to leave feedback about specification? Here, in this thread or on
>> another mailing list?
>>
>
> For the initial feedback please use the official JSR381 mailing list at
> https://groups.io/g/visrec
>
> Best
> Zoran
>
>
>>
>> Sincerely yours,
>>               Alexey Zinoviev
>>
>> ср, 1 апр. 2020 г. в 16:59, Zoran Sevarac <[hidden email]>:
>>
>>> On Tue, Mar 31, 2020 at 3:03 PM Alexey Zinoviev <[hidden email]>
>>> wrote:
>>>
>>>> Nice to see all of you guys (thanks Vladimir) here on dev-list of
>>>> Apache Ignite that includes distributed ML.
>>>>
>>>
>>> Nice to meet you too Alexey, and thanks to Vladimir for introducing us
>>> to this community.
>>>
>>>
>>>> First of all, want to say thank you for your great work to build specs
>>>> to the Java world, hope it will be helpful for Java devs in the nearest
>>>> future.
>>>> In the second, a few days ago I found a link
>>>> <https://twitter.com/frankgreco/status/1242885726196424712> in Frank
>>>> Greco twitter to this JSR and read a few examples to understand the
>>>> proposed API.
>>>> Also, we (in Apache Ignite ML) have experimental integration with
>>>> TensorFlow for inference and distributed training purposes and I'm
>>>> interested in TF Java API development  and DL4j/PyTorch possible
>>>> integration.
>>>> Will be great to have common API for different implementations.
>>>>
>>>
>>> Great, we're interested to hear about your experience with these libs,
>>> and requirements for distributed training. I also saw
>>> https://apacheignite.readme.io/docs/machine-learning  it looks like
>>> something different from Tensorflow integration? I guess it would make
>>> sense for end-users to use the same API for the same task regardless of the
>>> backend library.
>>> By the way, I recommend that you take a look at https://djl.ai it's
>>> deep learning library from Amazon which is a Java wrapper around MxNet, TF
>>> and PyTorch, they also participate in JSR381, and have provided initial
>>> implementation.
>>>
>>>
>>>> P.S. From my side, I've ready to join your team if the specification is
>>>> not finished yet, because I have a few comments and thoughts based on my
>>>> experience of different API usage. A few weeks ago I also joined to the
>>>> developer group of JVM API for TensorFlow (SIG JVM), hope it will be
>>>> helpful in both directions to communicate each other.
>>>>
>>>
>>> Sounds good,  at this phase we're looking for the input and feedback
>>> from various communities in order to make API that will meet their needs.
>>> So we would definitely like to hear your thoughts and we're open to
>>> adapt the specification.
>>>
>>> Best regards
>>> Zoran
>>>
>>>
>>>
>>>>
>>>> пн, 30 мар. 2020 г. в 21:00, Vladimir Ozerov <[hidden email]>:
>>>>
>>>>> Hi Alexey, Igniters,
>>>>>
>>>>> Let me introduce you Heather, Zoran, and Frank. Heather is the Chair
>>>>> of the JCP. Zoran and Frank are JSR 381 spec leads. They are interested in
>>>>> discussing the upcoming visual recognition specification with the Apache
>>>>> Ignite community, to understand whether the community has any interest in
>>>>> implementing it.
>>>>>
>>>>> Zoran and Frank,
>>>>>
>>>>> Please meet Alexey Zinoviev, who is the principal maintainer of the
>>>>> Apache ML module. I hope he will be able to help you with your questions.
>>>>>
>>>>> Regards,
>>>>> Vladimir.
>>>>>
>>>>
>>>
>>> --
>>> Zoran Sevarac, PhD, Associate Professor
>>> University of Belgrade, Faculty of Organisational Sciences, Department
>>> for Software Engineering
>>> Java Champion
>>> <https://community.oracle.com/community/java/java-champions> | Oracle
>>> Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
>>> Co-founder & CEO
>>>
>>
>
> --
> Zoran Sevarac, PhD, Associate Professor
> University of Belgrade, Faculty of Organisational Sciences, Department
> for Software Engineering
> Java Champion <https://community.oracle.com/community/java/java-champions>
> | Oracle Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
> Co-founder & CEO
>

[Zoran Sevarac]

Great, thanks, looking forward to it. In the meantime, I'll take a more
detailed look at the existing Apache Ignite ML support

Best,
Zoran

On Wed, Apr 1, 2020 at 7:59 PM Alexey Zinoviev <[hidden email]>
wrote:

> Thanks, I will check it out. Great dlj library with the reference JSR
> implementation, I have not seen yet this repo.
>
> Will return in a few weeks with suggestions on JSR mailing list
>
> Sincerely yours,
>    Alexey Zinoviev
>
> ср, 1 апр. 2020 г., 18:49 Zoran Sevarac <[hidden email]>:
>
>> I look on the https://djl.ai, it seems a little bit different with JSR
>>> (but close in builders), also as I understand (looking in sources) they
>>> don't support TF engine yet
>>> Is it a mature framework, how do you think?
>>>
>>
>> It is relatively new but very promising in my opinion. TF support is not
>> publically available but they do have it.
>> And also they provide implementation of JSR381 at
>> https://github.com/JavaVisRec/visrec-djl
>> They have recently joined JSR381
>>
>>
>>> Happy to join the unification movement in Java DL, where is the best
>>> place to leave feedback about specification? Here, in this thread or on
>>> another mailing list?
>>>
>>
>> For the initial feedback please use the official JSR381 mailing list at
>> https://groups.io/g/visrec
>>
>> Best
>> Zoran
>>
>>
>>>
>>> Sincerely yours,
>>>               Alexey Zinoviev
>>>
>>> ср, 1 апр. 2020 г. в 16:59, Zoran Sevarac <[hidden email]>:
>>>
>>>> On Tue, Mar 31, 2020 at 3:03 PM Alexey Zinoviev <[hidden email]>
>>>> wrote:
>>>>
>>>>> Nice to see all of you guys (thanks Vladimir) here on dev-list of
>>>>> Apache Ignite that includes distributed ML.
>>>>>
>>>>
>>>> Nice to meet you too Alexey, and thanks to Vladimir for introducing us
>>>> to this community.
>>>>
>>>>
>>>>> First of all, want to say thank you for your great work to build specs
>>>>> to the Java world, hope it will be helpful for Java devs in the nearest
>>>>> future.
>>>>> In the second, a few days ago I found a link
>>>>> <https://twitter.com/frankgreco/status/1242885726196424712> in Frank
>>>>> Greco twitter to this JSR and read a few examples to understand the
>>>>> proposed API.
>>>>> Also, we (in Apache Ignite ML) have experimental integration with
>>>>> TensorFlow for inference and distributed training purposes and I'm
>>>>> interested in TF Java API development  and DL4j/PyTorch possible
>>>>> integration.
>>>>> Will be great to have common API for different implementations.
>>>>>
>>>>
>>>> Great, we're interested to hear about your experience with these libs,
>>>> and requirements for distributed training. I also saw
>>>> https://apacheignite.readme.io/docs/machine-learning  it looks like
>>>> something different from Tensorflow integration? I guess it would make
>>>> sense for end-users to use the same API for the same task regardless of the
>>>> backend library.
>>>> By the way, I recommend that you take a look at https://djl.ai it's
>>>> deep learning library from Amazon which is a Java wrapper around MxNet, TF
>>>> and PyTorch, they also participate in JSR381, and have provided initial
>>>> implementation.
>>>>
>>>>
>>>>> P.S. From my side, I've ready to join your team if the specification
>>>>> is not finished yet, because I have a few comments and thoughts based on my
>>>>> experience of different API usage. A few weeks ago I also joined to the
>>>>> developer group of JVM API for TensorFlow (SIG JVM), hope it will be
>>>>> helpful in both directions to communicate each other.
>>>>>
>>>>
>>>> Sounds good,  at this phase we're looking for the input and feedback
>>>> from various communities in order to make API that will meet their needs.
>>>> So we would definitely like to hear your thoughts and we're open to
>>>> adapt the specification.
>>>>
>>>> Best regards
>>>> Zoran
>>>>
>>>>
>>>>
>>>>>
>>>>> пн, 30 мар. 2020 г. в 21:00, Vladimir Ozerov <[hidden email]>:
>>>>>
>>>>>> Hi Alexey, Igniters,
>>>>>>
>>>>>> Let me introduce you Heather, Zoran, and Frank. Heather is the Chair
>>>>>> of the JCP. Zoran and Frank are JSR 381 spec leads. They are interested in
>>>>>> discussing the upcoming visual recognition specification with the Apache
>>>>>> Ignite community, to understand whether the community has any interest in
>>>>>> implementing it.
>>>>>>
>>>>>> Zoran and Frank,
>>>>>>
>>>>>> Please meet Alexey Zinoviev, who is the principal maintainer of the
>>>>>> Apache ML module. I hope he will be able to help you with your questions.
>>>>>>
>>>>>> Regards,
>>>>>> Vladimir.
>>>>>>
>>>>>
>>>>
>>>> --
>>>> Zoran Sevarac, PhD, Associate Professor
>>>> University of Belgrade, Faculty of Organisational Sciences, Department
>>>> for Software Engineering
>>>> Java Champion
>>>> <https://community.oracle.com/community/java/java-champions> | Oracle
>>>> Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
>>>> Co-founder & CEO
>>>>
>>>
>>
>> --
>> Zoran Sevarac, PhD, Associate Professor
>> University of Belgrade, Faculty of Organisational Sciences, Department
>> for Software Engineering
>> Java Champion
>> <https://community.oracle.com/community/java/java-champions> | Oracle
>> Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
>> Co-founder & CEO
>>
>

--
Zoran Sevarac, PhD, Associate Professor
University of Belgrade, Faculty of Organisational Sciences, Department for
Software Engineering
Java Champion <https://community.oracle.com/community/java/java-champions>
| Oracle Groundbreaker Ambassador | Deep Netts <http://www.deepnetts.com>
Co-founder & CEO