[Theory] NOW: REMINDER: 2/17 Talks at TTIC: Amirali Aghazadeh, UC Berkeley

[Mary Marre]

*When:*      Wednesday, February 17th at* 11:10 am CT*



*Where:*     Zoom Virtual Talk (*register in advance here
<https://uchicagogroup.zoom.us/webinar/register/WN_aJxlZzpcSeWJqnh6Hjmq2w>*)



*Who: *       Amirali Aghazadeh, UC Berkeley


Title. Inferring Biological Functions with Explainable Algorithms


Abstract. Data-driven machine learning models that infer biological
functions from sequences are replacing the costly experimental measurements
in a number of application areas in biology including protein,
small-molecule, and genome engineering. These data-driven models largely
owe their success to the recent advancements in over-parameterized models
in machine learning such as deep neural networks (DNNs). However, the
number of labeled sequences available for training such models has remained
small compared to the vastness of the combinatorial sequence space. In
addition, these biological functions are typically complex, manifesting as
rugged landscapes with high-order epistatic (nonlinear) interactions. The
combination of these two factors makes the biological inference problem
statistically challenging.



In this talk, I view the problem of inferring biological functions from a
statistical signal processing perspective. I first discuss a fundamental
interpretation-computation tradeoff in explaining DNNs in terms of their
epistatic interactions. I then discuss how to develop a new hybrid
algorithm that blends techniques from optimization and coding theory to
regularize DNNs for inducing a biologically-relevant prior into their
architecture. Our combinatorial method enables DNNs to predict protein
functions using up to three times less number of sequences and explains
them in terms of their higher-order epistatic interactions.

Bio. Amirali Aghazadeh is a postdoctoral researcher in the Electrical
Engineering and Computer Science department at the University of
California, Berkeley, working with Kannan Ramchandran and Jennifer
Listgarten. Prior to that, he was a postdoctoral researcher at Stanford
University working with David Tse. He received his PhD degree in Electrical
and Computer Engineering from Rice University with Richard Baraniuk in
2017. His research interest is at the intersection of machine learning,
signal processing, inverse problems, and computational biology. He is the
recipient of the Hershel M. Rich Invention Award for his thesis on rapid
methods for DNA sensing as well as the Texas Instruments Fellowship for his
graduate studies. He received his Bachelor’s degree in Electrical
Engineering from Sharif University of Technology.

*Host:* Jinbo Xu <j3xu at ttic.edu>

Mary C. Marre
Faculty Administrative Support
*Toyota Technological Institute*
*6045 S. Kenwood Avenue*
*Room 517*
*Chicago, IL  60637*
*p:(773) 834-1757*
*f: (773) 357-6970*
*mmarre at ttic.edu <mmarre at ttic.edu>*


On Wed, Feb 17, 2021 at 10:31 AM Mary Marre <mmarre at ttic.edu> wrote:

> *When:*      Wednesday, February 17th at* 11:10 am CT*
>
>
>
> *Where:*     Zoom Virtual Talk (*register in advance here
> <https://uchicagogroup.zoom.us/webinar/register/WN_aJxlZzpcSeWJqnh6Hjmq2w>*
> )
>
>
>
> *Who: *       Amirali Aghazadeh, UC Berkeley
>
>
> Title. Inferring Biological Functions with Explainable Algorithms
>
>
> Abstract. Data-driven machine learning models that infer biological
> functions from sequences are replacing the costly experimental measurements
> in a number of application areas in biology including protein,
> small-molecule, and genome engineering. These data-driven models largely
> owe their success to the recent advancements in over-parameterized models
> in machine learning such as deep neural networks (DNNs). However, the
> number of labeled sequences available for training such models has remained
> small compared to the vastness of the combinatorial sequence space. In
> addition, these biological functions are typically complex, manifesting as
> rugged landscapes with high-order epistatic (nonlinear) interactions. The
> combination of these two factors makes the biological inference problem
> statistically challenging.
>
>
>
> In this talk, I view the problem of inferring biological functions from a
> statistical signal processing perspective. I first discuss a fundamental
> interpretation-computation tradeoff in explaining DNNs in terms of their
> epistatic interactions. I then discuss how to develop a new hybrid
> algorithm that blends techniques from optimization and coding theory to
> regularize DNNs for inducing a biologically-relevant prior into their
> architecture. Our combinatorial method enables DNNs to predict protein
> functions using up to three times less number of sequences and explains
> them in terms of their higher-order epistatic interactions.
>
> Bio. Amirali Aghazadeh is a postdoctoral researcher in the Electrical
> Engineering and Computer Science department at the University of
> California, Berkeley, working with Kannan Ramchandran and Jennifer
> Listgarten. Prior to that, he was a postdoctoral researcher at Stanford
> University working with David Tse. He received his PhD degree in Electrical
> and Computer Engineering from Rice University with Richard Baraniuk in
> 2017. His research interest is at the intersection of machine learning,
> signal processing, inverse problems, and computational biology. He is the
> recipient of the Hershel M. Rich Invention Award for his thesis on rapid
> methods for DNA sensing as well as the Texas Instruments Fellowship for his
> graduate studies. He received his Bachelor’s degree in Electrical
> Engineering from Sharif University of Technology.
>
> *Host:* Jinbo Xu <j3xu at ttic.edu>
>
> Mary C. Marre
> Faculty Administrative Support
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue*
> *Room 517*
> *Chicago, IL  60637*
> *p:(773) 834-1757*
> *f: (773) 357-6970*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
>
> On Tue, Feb 16, 2021 at 3:30 PM Mary Marre <mmarre at ttic.edu> wrote:
>
>> *When:*      Wednesday, February 17th at* 11:10 am CT*
>>
>>
>>
>> *Where:*     Zoom Virtual Talk (*register in advance here
>> <https://uchicagogroup.zoom.us/webinar/register/WN_aJxlZzpcSeWJqnh6Hjmq2w>*
>> )
>>
>>
>>
>> *Who: *       Amirali Aghazadeh, UC Berkeley
>>
>>
>> Title. Inferring Biological Functions with Explainable Algorithms
>>
>>
>> Abstract. Data-driven machine learning models that infer biological
>> functions from sequences are replacing the costly experimental measurements
>> in a number of application areas in biology including protein,
>> small-molecule, and genome engineering. These data-driven models largely
>> owe their success to the recent advancements in over-parameterized models
>> in machine learning such as deep neural networks (DNNs). However, the
>> number of labeled sequences available for training such models has remained
>> small compared to the vastness of the combinatorial sequence space. In
>> addition, these biological functions are typically complex, manifesting as
>> rugged landscapes with high-order epistatic (nonlinear) interactions. The
>> combination of these two factors makes the biological inference problem
>> statistically challenging.
>>
>>
>>
>> In this talk, I view the problem of inferring biological functions from a
>> statistical signal processing perspective. I first discuss a fundamental
>> interpretation-computation tradeoff in explaining DNNs in terms of their
>> epistatic interactions. I then discuss how to develop a new hybrid
>> algorithm that blends techniques from optimization and coding theory to
>> regularize DNNs for inducing a biologically-relevant prior into their
>> architecture. Our combinatorial method enables DNNs to predict protein
>> functions using up to three times less number of sequences and explains
>> them in terms of their higher-order epistatic interactions.
>>
>> Bio. Amirali Aghazadeh is a postdoctoral researcher in the Electrical
>> Engineering and Computer Science department at the University of
>> California, Berkeley, working with Kannan Ramchandran and Jennifer
>> Listgarten. Prior to that, he was a postdoctoral researcher at Stanford
>> University working with David Tse. He received his PhD degree in Electrical
>> and Computer Engineering from Rice University with Richard Baraniuk in
>> 2017. His research interest is at the intersection of machine learning,
>> signal processing, inverse problems, and computational biology. He is the
>> recipient of the Hershel M. Rich Invention Award for his thesis on rapid
>> methods for DNA sensing as well as the Texas Instruments Fellowship for his
>> graduate studies. He received his Bachelor’s degree in Electrical
>> Engineering from Sharif University of Technology.
>>
>> *Host:* Jinbo Xu <j3xu at ttic.edu>
>>
>>
>>
>> Mary C. Marre
>> Faculty Administrative Support
>> *Toyota Technological Institute*
>> *6045 S. Kenwood Avenue*
>> *Room 517*
>> *Chicago, IL  60637*
>> *p:(773) 834-1757*
>> *f: (773) 357-6970*
>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>
>>
>> On Wed, Feb 10, 2021 at 10:09 PM Mary Marre <mmarre at ttic.edu> wrote:
>>
>>> *When:*      Wednesday, February 17th at* 11:10 am CT*
>>>
>>>
>>>
>>> *Where:*     Zoom Virtual Talk (*register in advance here
>>> <https://uchicagogroup.zoom.us/webinar/register/WN_aJxlZzpcSeWJqnh6Hjmq2w>*
>>> )
>>>
>>>
>>>
>>> *Who: *       Amirali Aghazadeh, UC Berkeley
>>>
>>>
>>> Title. Inferring Biological Functions with Explainable Algorithms
>>>
>>> Abstract. Data-driven machine learning models that infer biological
>>> functions from sequences are replacing the costly experimental measurements
>>> in a number of application areas in biology including protein,
>>> small-molecule, and genome engineering. These data-driven models largely
>>> owe their success to the recent advancements in over-parameterized models
>>> in machine learning such as deep neural networks (DNNs). However, the
>>> number of labeled sequences available for training such models has remained
>>> small compared to the vastness of the combinatorial sequence space. In
>>> addition, these biological functions are typically complex, manifesting as
>>> rugged landscapes with high-order epistatic (nonlinear) interactions. The
>>> combination of these two factors makes the biological inference problem
>>> statistically challenging.
>>>
>>> In this talk, I view the problem of inferring biological functions from
>>> a statistical signal processing perspective. I first discuss a fundamental
>>> interpretation-computation tradeoff in explaining DNNs in terms of their
>>> epistatic interactions. I then discuss how to develop a new hybrid
>>> algorithm that blends techniques from optimization and coding theory to
>>> regularize DNNs for inducing a biologically-relevant prior into their
>>> architecture. Our combinatorial method enables DNNs to predict protein
>>> functions using up to three times less number of sequences and explains
>>> them in terms of their higher-order epistatic interactions.
>>>
>>> Bio. Amirali Aghazadeh is a postdoctoral researcher in the Electrical
>>> Engineering and Computer Science department at the University of
>>> California, Berkeley, working with Kannan Ramchandran and Jennifer
>>> Listgarten. Prior to that, he was a postdoctoral researcher at Stanford
>>> University working with David Tse. He received his PhD degree in Electrical
>>> and Computer Engineering from Rice University with Richard Baraniuk in
>>> 2017. His research interest is at the intersection of machine learning,
>>> signal processing, inverse problems, and computational biology. He is the
>>> recipient of the Hershel M. Rich Invention Award for his thesis on rapid
>>> methods for DNA sensing as well as the Texas Instruments Fellowship for his
>>> graduate studies. He received his Bachelor’s degree in Electrical
>>> Engineering from Sharif University of Technology.
>>>
>>> *Host:* Jinbo Xu <j3xu at ttic.edu>
>>>
>>>
>>> Mary C. Marre
>>> Faculty Administrative Support
>>> *Toyota Technological Institute*
>>> *6045 S. Kenwood Avenue*
>>> *Room 517*
>>> *Chicago, IL  60637*
>>> *p:(773) 834-1757*
>>> *f: (773) 357-6970*
>>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20210217/852597a4/attachment-0001.html>