[Theory] 2/10 Talks at TTIC: Tanya Goyal, University of Texas at Austin

Mary Marre mmarre at ttic.edu
Fri Feb 3 15:00:00 CST 2023


*When:*        Friday, February 10th at* 11:00** a**m CT   *


*Where:       *Talk will be given *live, in-person* at

                   TTIC, 6045 S. Kenwood Avenue

                   5th Floor, Room 530


*Virtually:*  *via* Panopto (*livestream
<https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=d071b447-00c1-4271-8fe4-af9501376e56>*
)


*Who: *        Tanya Goyal, University of Texas at Austin


------------------------------
*Title:*          Building Reliable Text Generation Capabilities for Large
Language Models

*Abstract:* Recent breakthroughs in NLP, e.g. GPT-3, have equipped AI
systems with impressive text generation capabilities that have the
potential to transform how we interact with machines. However, despite this
success, these models still suffer from critical limitations that stand in
the way of widespread adoption, such as generating factually incorrect
information. In my talk, I will describe my work that addresses these
limitations. First, I will describe my work on building evaluation tools to
detect errors in model generated text along critical dimensions like
factuality by grounding evaluation in actual error distributions. Then, I
will describe training techniques that target such limitations and produce
models that generate higher quality text. Ultimately, progress along both
these is required to deliver reliable systems that work beyond standard
benchmarks in real world systems.

*Bio:* Tanya Goyal is currently a Ph.D student at the University of Texas
at Austin, advised by Prof. Greg Durrett. She did her undergraduate at the
Indian Institute of Technology, Guwahati and worked at Adobe Research
before graduate school. Her research focuses on making text generation
models more reliable, building both better training techniques to improve
quality as well as evaluation tools to better benchmark their performance.

*Host:* Karen Livescu <klivescu at ttic.edu>


Mary C. Marre
Faculty Administrative Support
*Toyota Technological Institute*
*6045 S. Kenwood Avenue, Rm 517*
*Chicago, IL  60637*
*773-834-1757*
*mmarre at ttic.edu <mmarre at ttic.edu>*


On Tue, Jan 31, 2023 at 8:16 PM Mary Marre <mmarre at ttic.edu> wrote:

> *The 2/1 Talk is RESCHEDULED to Friday, 2/10*
>
> *When:*        New date/time is Friday, February 10th at* 11:00** a**m
> CT   *
>
>
> *Where:       *Talk will be given *live, in-person* at
>
>                    TTIC, 6045 S. Kenwood Avenue
>
>                    5th Floor, Room 530
>
>
> *Virtually:*  *via* Panopto (*livestream
> <https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=d071b447-00c1-4271-8fe4-af9501376e56>*
> )
>
>
> *Who: *        Tanya Goyal, University of Texas at Austin
>
>
> ------------------------------
> *Title:*          Building Reliable Text Generation Capabilities for
> Large Language Models
>
> *Abstract:* Recent breakthroughs in NLP, e.g. GPT-3, have equipped AI
> systems with impressive text generation capabilities that have the
> potential to transform how we interact with machines. However, despite this
> success, these models still suffer from critical limitations that stand in
> the way of widespread adoption, such as generating factually incorrect
> information. In my talk, I will describe my work that addresses these
> limitations. First, I will describe my work on building evaluation tools to
> detect errors in model generated text along critical dimensions like
> factuality by grounding evaluation in actual error distributions. Then, I
> will describe training techniques that target such limitations and produce
> models that generate higher quality text. Ultimately, progress along both
> these is required to deliver reliable systems that work beyond standard
> benchmarks in real world systems.
>
> *Bio:* Tanya Goyal is currently a Ph.D student at the University of Texas
> at Austin, advised by Prof. Greg Durrett. She did her undergraduate at the
> Indian Institute of Technology, Guwahati and worked at Adobe Research
> before graduate school. Her research focuses on making text generation
> models more reliable, building both better training techniques to improve
> quality as well as evaluation tools to better benchmark their performance.
>
> *Host:* Karen Livescu <klivescu at ttic.edu>
>
>
> Mary C. Marre
> Faculty Administrative Support
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue, Rm 517*
> *Chicago, IL  60637*
> *773-834-1757*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
>
> On Tue, Jan 31, 2023 at 2:58 PM Mary Marre <mmarre at ttic.edu> wrote:
>
>> *When:*        Wednesday, February 1st at* 11:30** a**m CT   *
>>
>>
>> *Where:       *Talk will be given *live, in-person* at
>>
>>                    TTIC, 6045 S. Kenwood Avenue
>>
>>                    5th Floor, Room 530
>>
>>
>> *Virtually:*  *via* Panopto (*livestream
>> <https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=d071b447-00c1-4271-8fe4-af9501376e56>*
>> )
>>
>>
>> *Who: *        Tanya Goyal, University of Texas at Austin
>>
>>
>> ------------------------------
>> *Title:*  Building Reliable Text Generation Capabilities for Large
>> Language Models
>>
>> *Abstract:* Recent breakthroughs in NLP, e.g. GPT-3, have equipped AI
>> systems with impressive text generation capabilities that have the
>> potential to transform how we interact with machines. However, despite this
>> success, these models still suffer from critical limitations that stand in
>> the way of widespread adoption, such as generating factually incorrect
>> information. In my talk, I will describe my work that addresses these
>> limitations. First, I will describe my work on building evaluation tools to
>> detect errors in model generated text along critical dimensions like
>> factuality by grounding evaluation in actual error distributions. Then, I
>> will describe training techniques that target such limitations and produce
>> models that generate higher quality text. Ultimately, progress along both
>> these is required to deliver reliable systems that work beyond standard
>> benchmarks in real world systems.
>>
>> *Bio:* Tanya Goyal is currently a Ph.D student at the University of
>> Texas at Austin, advised by Prof. Greg Durrett. She did her undergraduate
>> at the Indian Institute of Technology, Guwahati and worked at Adobe
>> Research before graduate school. Her research focuses on making text
>> generation models more reliable, building both better training techniques
>> to improve quality as well as evaluation tools to better benchmark their
>> performance.
>>
>> *Host:* Karen Livescu <klivescu at ttic.edu>
>>
>>
>>
>>
>> Mary C. Marre
>> Faculty Administrative Support
>> *Toyota Technological Institute*
>> *6045 S. Kenwood Avenue, Rm 517*
>> *Chicago, IL  60637*
>> *773-834-1757*
>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>
>>
>> On Wed, Jan 25, 2023 at 1:20 PM Mary Marre <mmarre at ttic.edu> wrote:
>>
>>> *When:*        Wednesday, February 1st at* 11:30** a**m CT   *
>>>
>>>
>>> *Where:       *Talk will be given *live, in-person* at
>>>
>>>                    TTIC, 6045 S. Kenwood Avenue
>>>
>>>                    5th Floor, Room 530
>>>
>>>
>>> *Virtually:*  *via* Panopto (*livestream
>>> <https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=d071b447-00c1-4271-8fe4-af9501376e56>*
>>> )
>>>
>>>
>>> *Who: *        Tanya Goyal, University of Texas at Austin
>>>
>>>
>>> ------------------------------
>>> *Title:*  Building Reliable Text Generation Capabilities for Large
>>> Language Models
>>>
>>> *Abstract:* Recent breakthroughs in NLP, e.g. GPT-3, have equipped AI
>>> systems with impressive text generation capabilities that have the
>>> potential to transform how we interact with machines. However, despite this
>>> success, these models still suffer from critical limitations that stand in
>>> the way of widespread adoption, such as generating factually incorrect
>>> information. In my talk, I will describe my work that addresses these
>>> limitations. First, I will describe my work on building evaluation tools to
>>> detect errors in model generated text along critical dimensions like
>>> factuality by grounding evaluation in actual error distributions. Then, I
>>> will describe training techniques that target such limitations and produce
>>> models that generate higher quality text. Ultimately, progress along both
>>> these is required to deliver reliable systems that work beyond standard
>>> benchmarks in real world systems.
>>>
>>> *Bio:* Tanya Goyal is currently a Ph.D student at the University of
>>> Texas at Austin, advised by Prof. Greg Durrett. She did her undergraduate
>>> at the Indian Institute of Technology, Guwahati and worked at Adobe
>>> Research before graduate school. Her research focuses on making text
>>> generation models more reliable, building both better training techniques
>>> to improve quality as well as evaluation tools to better benchmark their
>>> performance.
>>>
>>> *Host:* Karen Livescu <klivescu at ttic.edu>
>>>
>>>
>>> Mary C. Marre
>>> Faculty Administrative Support
>>> *Toyota Technological Institute*
>>> *6045 S. Kenwood Avenue, Rm 517*
>>> *Chicago, IL  60637*
>>> *773-834-1757*
>>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20230203/0c46712f/attachment-0001.html>


More information about the Theory mailing list