From predragp at andrew.cmu.edu Thu Dec 1 23:14:49 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Thu, 1 Dec 2022 23:14:49 -0500 Subject: GPUs 1-9 offline In-Reply-To: References: Message-ID: Hi Conor, I just noticed myself. It is not just GPUs 1-9 it is also Denver. The common thing for all those 10 servers is that they draw electricity from the same Metered 17.3 kW PDU. Sure enough IPMI is off as well which confirms that there is no electric power in that server RACK. Somebody cut the electricity to the RACK A1-2A or PDU had a catastrophic failure. I am now calling the server room to have them physically inspect the rack. Best, Predrag On Thu, Dec 1, 2022 at 6:37 PM Conor Igoe wrote: > Predrag, > > Sorry to bother you, but I was wondering if you knew why GPUs 1-9 are > offline since earlier today? > > Best, > *Conor* > -------------- next part -------------- An HTML attachment was scrubbed... URL: From predragp at andrew.cmu.edu Thu Dec 1 23:38:30 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Thu, 1 Dec 2022 23:38:30 -0500 Subject: GPUs 1-9 offline In-Reply-To: References:

Message-ID: PDU's master cable plug (IEC 60309 60 A 3P + PE) is still plugged into the floor outlet. Somebody was either messing with the main electric switchboard or we have catastrophic failure of the PDU. I emailed the director of CS computing facility Ed Walter. I would be very surprised that any major electric work (switchboard) was done without me not knowing about it. CMU doesn't have people who are licensed to do that kind of work. We hire external crew and such for is scheduled months in advance. I will inspect the cables and the unit tomorrow after I hear back from Ed. It looks like the replacement PDU is close to $4000. We used to buy them for about $1800. Best, Predrag On Thu, Dec 1, 2022 at 11:14 PM Predrag Punosevac wrote: > Hi Conor, > > I just noticed myself. It is not just GPUs 1-9 it is also Denver. The > common thing for all those 10 servers is that they draw electricity from > the same Metered 17.3 kW PDU. Sure enough IPMI is off as well which > confirms that there is no electric power in that server RACK. Somebody cut > the electricity to the RACK A1-2A or PDU had a catastrophic failure. I am > now calling the server room to have them physically inspect the rack. > > Best, > Predrag > > On Thu, Dec 1, 2022 at 6:37 PM Conor Igoe wrote: > >> Predrag, >> >> Sorry to bother you, but I was wondering if you knew why GPUs 1-9 are >> offline since earlier today? >> >> Best, >> *Conor* >> > -------------- next part -------------- An HTML attachment was scrubbed... URL: From predragp at andrew.cmu.edu Fri Dec 2 18:31:11 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Fri, 2 Dec 2022 18:31:11 -0500 Subject: GPUs 1-9 offline In-Reply-To: References:

Message-ID: Sorry for the delayed update. There is nothing wrong with electricity. The network switch is busted! I have two spare switches in the storage room. I am replacing it right now. Hopefully one of them will be good. Best, Predrag On Thu, Dec 1, 2022 at 11:38 PM Predrag Punosevac wrote: > PDU's master cable plug (IEC 60309 60 A 3P + PE) is still plugged into the > floor outlet. Somebody was either messing with the main electric > switchboard or we have catastrophic failure of the PDU. > > I emailed the director of CS computing facility Ed Walter. I would be very > surprised that any major electric work (switchboard) was done without me > not knowing about it. CMU doesn't have people who are licensed to do that > kind of work. We hire external crew and such for is scheduled months in > advance. > > I will inspect the cables and the unit tomorrow after I hear back from > Ed. It looks like the replacement PDU is close to $4000. We used to buy > them for about $1800. > > Best, > Predrag > > On Thu, Dec 1, 2022 at 11:14 PM Predrag Punosevac > wrote: > >> Hi Conor, >> >> I just noticed myself. It is not just GPUs 1-9 it is also Denver. The >> common thing for all those 10 servers is that they draw electricity from >> the same Metered 17.3 kW PDU. Sure enough IPMI is off as well which >> confirms that there is no electric power in that server RACK. Somebody cut >> the electricity to the RACK A1-2A or PDU had a catastrophic failure. I am >> now calling the server room to have them physically inspect the rack. >> >> Best, >> Predrag >> >> On Thu, Dec 1, 2022 at 6:37 PM Conor Igoe wrote: >> >>> Predrag, >>> >>> Sorry to bother you, but I was wondering if you knew why GPUs 1-9 are >>> offline since earlier today? >>> >>> Best, >>> *Conor* >>> >> -------------- next part -------------- An HTML attachment was scrubbed... URL: From predragp at andrew.cmu.edu Fri Dec 2 19:54:24 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Fri, 2 Dec 2022 19:54:24 -0500 Subject: GPUs 1-9 offline In-Reply-To: References:

Message-ID: Fixed! All computing nodes are available. Predrag On Fri, Dec 2, 2022 at 6:31 PM Predrag Punosevac wrote: > Sorry for the delayed update. There is nothing wrong with electricity. The > network switch is busted! I have two spare switches in the storage room. I > am replacing it right now. Hopefully one of them will be good. > > Best, > Predrag > > On Thu, Dec 1, 2022 at 11:38 PM Predrag Punosevac > wrote: > >> PDU's master cable plug (IEC 60309 60 A 3P + PE) is still plugged into >> the floor outlet. Somebody was either messing with the main electric >> switchboard or we have catastrophic failure of the PDU. >> >> I emailed the director of CS computing facility Ed Walter. I would be >> very surprised that any major electric work (switchboard) was done without >> me not knowing about it. CMU doesn't have people who are licensed to do >> that kind of work. We hire external crew and such for is scheduled months >> in advance. >> >> I will inspect the cables and the unit tomorrow after I hear back from >> Ed. It looks like the replacement PDU is close to $4000. We used to buy >> them for about $1800. >> >> Best, >> Predrag >> >> On Thu, Dec 1, 2022 at 11:14 PM Predrag Punosevac < >> predragp at andrew.cmu.edu> wrote: >> >>> Hi Conor, >>> >>> I just noticed myself. It is not just GPUs 1-9 it is also Denver. The >>> common thing for all those 10 servers is that they draw electricity from >>> the same Metered 17.3 kW PDU. Sure enough IPMI is off as well which >>> confirms that there is no electric power in that server RACK. Somebody cut >>> the electricity to the RACK A1-2A or PDU had a catastrophic failure. I am >>> now calling the server room to have them physically inspect the rack. >>> >>> Best, >>> Predrag >>> >>> On Thu, Dec 1, 2022 at 6:37 PM Conor Igoe wrote: >>> >>>> Predrag, >>>> >>>> Sorry to bother you, but I was wondering if you knew why GPUs 1-9 are >>>> offline since earlier today? >>>> >>>> Best, >>>> *Conor* >>>> >>> -------------- next part -------------- An HTML attachment was scrubbed... URL: From awd at cs.cmu.edu Sun Dec 4 20:27:55 2022 From: awd at cs.cmu.edu (Artur Dubrawski) Date: Sun, 4 Dec 2022 20:27:55 -0500 Subject: Autonian on the prestigious Forbes 30 under 30 list Message-ID: Team, Please check this out: https://www.forbes.com/profile/eric-lei/?sh=62cb11964e16 and join me in congratulating Eric on this massive accomplishment! Cheers, Artur PS I have to admit I teased Eric a lot about him switching to the dark side when upon graduation he decided to pursue a career in fintech, but now I feel that maybe I should reconsider. Not that I've just changed my fundamental views on the topic, but the Auton Lab could certainly use more office space, and maybe some hedge fund superstar could help us build it. Heck, we could even call it "Lei Hall" or something... :) -------------- next part -------------- An HTML attachment was scrubbed... URL: From roni at cmu.edu Sun Dec 4 20:46:38 2022 From: roni at cmu.edu (Roni Rosenfeld) Date: Sun, 4 Dec 2022 20:46:38 -0500 Subject: Autonian on the prestigious Forbes 30 under 30 list In-Reply-To: References: Message-ID: Congrats indeed, to Eric and to the Auton Lab! Roni --- Roni Rosenfeld , Professor and Head, Machine Learning Department, School of Computer Science, Carnegie Mellon University On Sun, Dec 4, 2022 at 8:28 PM Artur Dubrawski wrote: > Team, > > Please check this out: > > https://www.forbes.com/profile/eric-lei/?sh=62cb11964e16 > > and join me in congratulating Eric on this massive accomplishment! > > Cheers, > Artur > > PS I have to admit I teased Eric a lot about him switching to the dark > side when upon graduation he decided to pursue a career in fintech, but now > I feel that maybe I should reconsider. Not that I've just changed my > fundamental views on the topic, but the Auton Lab could certainly use more > office space, and maybe some hedge fund superstar could help us build it. > Heck, we could even call it "Lei Hall" or something... :) > -------------- next part -------------- An HTML attachment was scrubbed... URL: From stidle at andrew.cmu.edu Mon Dec 5 09:25:30 2022 From: stidle at andrew.cmu.edu (Diane Stidle) Date: Mon, 5 Dec 2022 09:25:30 -0500 Subject: Autonian on the prestigious Forbes 30 under 30 list In-Reply-To: References: Message-ID: <57a43b65-0e6e-f9a3-8db1-80a37c664b70@andrew.cmu.edu> Hi Artur, Thanks for letting us know about this! Eric was just at our NeurIPS party and didn't say anything about it! Best, Diane On 12/4/2022 8:27 PM, Artur Dubrawski wrote: > Team, > > Please check this out: > > https://www.forbes.com/profile/eric-lei/?sh=62cb11964e16 > > and join me in congratulating Eric on this massive accomplishment! > > Cheers, > Artur > > PS I have to admit I teased Eric a lot about him switching?to the dark > side when upon graduation he decided to pursue a career in fintech, > but now I feel that maybe I should reconsider. Not that I've just > changed my fundamental views on the topic, but the Auton Lab?could > certainly use more office space, and maybe some hedge fund superstar > could help us build it. Heck, we could even call it "Lei Hall" or > something... :) From aaupperlee at cmu.edu Mon Dec 5 09:36:10 2022 From: aaupperlee at cmu.edu (Aaron Aupperlee) Date: Mon, 5 Dec 2022 09:36:10 -0500 Subject: Autonian on the prestigious Forbes 30 under 30 list In-Reply-To: <57a43b65-0e6e-f9a3-8db1-80a37c664b70@andrew.cmu.edu> References: <57a43b65-0e6e-f9a3-8db1-80a37c664b70@andrew.cmu.edu> Message-ID: Congrats! Adding Adam here. He tweeted about this last week on MLD's Twitter account. Aaron On Mon, Dec 5, 2022 at 9:25 AM Diane Stidle wrote: > Hi Artur, > > Thanks for letting us know about this! Eric was just at our NeurIPS > party and didn't say anything about it! > > Best, > Diane > > On 12/4/2022 8:27 PM, Artur Dubrawski wrote: > > Team, > > > > Please check this out: > > > > https://www.forbes.com/profile/eric-lei/?sh=62cb11964e16 > > > > and join me in congratulating Eric on this massive accomplishment! > > > > Cheers, > > Artur > > > > PS I have to admit I teased Eric a lot about him switching to the dark > > side when upon graduation he decided to pursue a career in fintech, > > but now I feel that maybe I should reconsider. Not that I've just > > changed my fundamental views on the topic, but the Auton Lab could > > certainly use more office space, and maybe some hedge fund superstar > > could help us build it. Heck, we could even call it "Lei Hall" or > > something... :) > -------------- next part -------------- An HTML attachment was scrubbed... URL: From awd at cs.cmu.edu Mon Dec 5 14:59:28 2022 From: awd at cs.cmu.edu (Artur Dubrawski) Date: Mon, 5 Dec 2022 14:59:28 -0500 Subject: Fwd: Thesis Proposal - Dec. 16, 2022 - Sebastian Caldas - Collaborative learning by leveraging siloed data In-Reply-To: References: Message-ID: An important and ingenious talk ahead of us! Artur ---------- Forwarded message --------- From: Diane Stidle Date: Mon, Dec 5, 2022 at 2:33 PM Subject: Thesis Proposal - Dec. 16, 2022 - Sebastian Caldas - Collaborative learning by leveraging siloed data To: ml-seminar at cs.cmu.edu , , < martin.jaggi at epfl.ch> *Thesis Proposal* Date: December 16, 2022 Time: 10:30am (EST) Remote Only Speaker: Sebastian Caldas *Title: Collaborative learning by leveraging siloed data* Abstract: Data holders cannot always share the data that they own, which can ultimately limit the modeling capabilities of each holder. For example, a hospital may lack representative records to learn about a new or rare condition, or a single mobile device may not have enough input to train a useful language model about its user. In both of these cases, these siloed data holders would benefit from collaborating with others in order to leverage their data. In recent years, the field of federated learning has taken an interest in learning performant collaborative models from siloed data. For these models to be truly useful, however, they must provide utility along dimensions beyond predictive performance, such as confidentiality, fairness and privacy. In this thesis proposal, I will demonstrate how to improve the utility of collaborative models that leverage siloed data, focusing on three dimensions of utility that are of current relevance to collaborative contexts: Explanations: We combine explanations with predictive performance in pursuit of true clinical utility. To this end, we introduced FRCLS, an algorithm that can explicitly identify when a prediction is using knowledge from an external collaborator, and provides interpretable rules that delineate subpopulations for which that external knowledge is useful. We have demonstrated the efficacy of FRCLS on a variety of clinical tasks including early prediction of sepsis and prediction of overly long lengths of stay. Expert supervision: We encode domain knowledge into on-device data, enabling collaborative learning for a wider variety of problems. We encode this knowledge by leveraging heuristics curated by experts. We first learn which heuristics will be useful for the devices? data and then train a weakly supervised federated model using these heuristics. Communication constraints: To complete my dissertation, I propose to study settings where collaborators are limited in the number of rounds of communication that can be exchanged, as is seen in clinical settings with limited infrastructure. I propose to develop an adaptive knowledge distillation strategy and to demonstrate it in a healthcare application context. *Thesis Committee:* Artur Dubrawski (chair) Virginia Smith Gilles Clermont (University of Pittsburgh) Martin Jaggi (EPFL) *Zoom meeting link:* https://cmu.zoom.us/j/94957077221?pwd=SkdHZitvNkR2Zm9lSXMyUGtPUldjQT09 *Link to the draft document: * https://drive.google.com/file/d/1Wu_ysaVm22G5PgOpTJbHy14DJ-uvr34U/view?usp=sharing -------------- next part -------------- An HTML attachment was scrubbed... URL: From awd at cs.cmu.edu Wed Dec 7 08:55:35 2022 From: awd at cs.cmu.edu (Artur Dubrawski) Date: Wed, 7 Dec 2022 08:55:35 -0500 Subject: REMINDER: RI Ph.D. Thesis Defense: Benedikt Boecking In-Reply-To: References: Message-ID: This is happening today at 1pm in a hybrid mode (Ben will be there in person). See you there! Artur On Mon, Nov 28, 2022 at 12:28 PM Artur Dubrawski wrote: > Please join Ben on his Big Day and see a really cool talk he is going to > give. > > Cheers > Artur > > > ---------- Forwarded message --------- > From: Suzanne Muth > Date: Mon, Nov 28, 2022 at 11:54 AM > Subject: RI Ph.D. Thesis Defense: Benedikt Boecking > To: > > > Date: 07 December 2022 > Time: 1:00 p.m. (ET) > Location: NSH 4305 > Zoom Link: > https://cmu.zoom.us/j/96368686155?pwd=Zm9abDRRYWNJUkNqU2pIZmEvM0hpQT09 > Type: Ph.D. Thesis Defense > Who: Benedikt Boecking > Title: Learning with Diverse Forms of Imperfect and Indirect Supervision > > Abstract: > Powerful Machine Learning (ML) models trained on large, annotated datasets > have driven impressive advances in fields including natural language > processing and computer vision. In turn, such developments have led to > impactful applications of ML in areas such as healthcare, e-commerce, and > predictive maintenance. However, obtaining annotated datasets at the scale > required for training high capacity ML models is frequently a bottleneck > for promising applications of ML. In this thesis, I study alternative > pathways for acquiring domain knowledge and develop methodologies to enable > learning from weak supervision, i.e., imperfect and indirect forms of > supervision. I cover three forms of weak supervision: pairwise linkage > feedback, programmatic weak supervision, and paired multi-modal data. These > forms of information are often easy to obtain at scale, and the methods I > develop reduce--and in some cases eliminate--the need for pointillistic > ground truth annotations. > > I begin by studying the utility of pairwise supervision. I introduce a new > constrained clustering method which uses small amounts of > pairwise constraints to simultaneously learn a kernel and cluster data. The > method outperforms related approaches on a large and diverse group > of publicly available datasets. Next, I introduce imperfect pairwise > supervision to programmatic weak supervision label models. I > show empirically that just one source of weak pairwise feedback can lead to > significantly improved downstream performance. > > I then further the study of programmatic data labeling methods by > introducing approaches that model the distribution of inputs in concert > with weak labels. I first introduce a framework for joint learning of a > label and end model on the basis of observed weak labels, > showing improvements over prior work in terms of end model performance on > downstream test sets. Next, I introduce a method that fuses > generative adversarial networks and programmatic weak supervision label > models to the benefit of both, measured by label model performance and > data generation quality. > > In the last part of this thesis, I tackle a central challenge in > programmatic weak supervision: the need for experts to provide labeling > rules. First, I introduce an interactive learning framework that aids users > in discovering weak supervision sources to capture subject matter > experts? knowledge of the application domain in an efficient fashion. I > then study the opportunity of dispensing with labeling functions altogether > by learning from unstructured natural language descriptions directly. In > particular, I study how biomedical text paired with images can be exploited > for self-supervised vision--language processing, yielding data-efficient > representations and enabling zero-shot classification, without requiring > experts to define rules on the text or images. > > Together, these works provide novel methodologies and frameworks to encode > and use expert domain knowledge more efficiently in ML models, reducing the > bottleneck created by the need for manual ground truth annotations. > > Thesis Committee Members: > Artur Dubrawski, Chair > Jeff Schneider > Barnabas Poczos > Hoifung Poon, Microsoft Research > > A draft of the thesis defense document is available at: > > https://drive.google.com/file/d/17DB_6gkfH7LPVzkt0adS0-O58pg_RSmE/view?usp=sharing > > _______________________________________________ > ri-people mailing list > ri-people at lists.andrew.cmu.edu > https://lists.andrew.cmu.edu/mailman/listinfo/ri-people > -------------- next part -------------- An HTML attachment was scrubbed... URL: From predragp at andrew.cmu.edu Fri Dec 9 17:46:33 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Fri, 9 Dec 2022 17:46:33 -0500 Subject: LOV 4 and LOV8 down Message-ID: Dear Autonians, Lov8 is refusing to reboot due to the faulty HDD used for scratch directories. I am at CMU right now but not sure if I will be able to fix it tonight. LOV4 is down reported without electricity. LOV4 is one of the oldest servers in the lab originally provisioned by Kyle Miller before I came to the Lab. Best case scenario for LOV4 is a dead power supply which is 5 minutes fix. If it is a dead motherboard the server will have to be trashed as it is older than 10 years. Best, Predrag -------------- next part -------------- An HTML attachment was scrubbed... URL: From awd at cs.cmu.edu Tue Dec 13 13:01:33 2022 From: awd at cs.cmu.edu (Artur Dubrawski) Date: Tue, 13 Dec 2022 13:01:33 -0500 Subject: Brief meetings with Artur this week In-Reply-To: References:

Message-ID: I've just published some openings for this week. This week turned out to be quite out of sync for me and so I can only open time on Fri and Sat. I hope it will work for most of us. Next week will be my chemo treatment week and I can't predict my availability yet. I would really need to meet with all PhD and MS student researchers before all the winter breaks. Thanks Artur > >> >> On Mon, Nov 7, 2022, 4:35 PM Artur Dubrawski wrote: >> >>> The short meeting slots for this week have just been opened. Please book >>> one (or more) while they last. >>> >>> Spreadsheet for signups link and the zoom link have not changed, they >>> are below for easy reference. >>> >>> Cheers >>> Artur >>> >>> >>>> >>>>>>> >>>>>>>> >>>>>>>> https://docs.google.com/spreadsheets/d/1OpY1DSxG7LLsMRroocMFTgqndSMiRTYUhYEwYdXH7Wc/edit?pli=1#gid=0 >>>>>>>> >>>>>>>> We will be using the same zoom link as before: >>>>>>>> >>>>>>>> https://cmu.zoom.us/j/9672166543 >>>>>>>> >>>>>>>> PS Let me know if the available times do not work for you and we >>>>>>>> will >>>>>>>> look for alternatives. >>>>>>>> >>>>>>> -------------- next part -------------- An HTML attachment was scrubbed... URL: From predragp at andrew.cmu.edu Tue Dec 13 14:38:00 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Tue, 13 Dec 2022 14:38:00 -0500 Subject: Unable to Access GPU1 In-Reply-To: References: Message-ID: Someone used /tmp to do Python cashing. Eventually the root ran out of space. The server had to be rebooted. Server is again available but the work is probably lost. Best, Predrag On Tue, Dec 13, 2022 at 12:13 PM Ian Char wrote: > Hey Predrag, > > Hope that you are doing well! > > It seems that GPU1 cannot be accessed anymore, and I was wondering if you > knew why. > > Thank you, > Ian > -------------- next part -------------- An HTML attachment was scrubbed... URL: From predragp at andrew.cmu.edu Tue Dec 13 22:00:02 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Tue, 13 Dec 2022 22:00:02 -0500 Subject: LOV 4 and LOV8 down In-Reply-To: References: Message-ID: I finally made it to CMU and fixed both servers tonight. LOV4 had a dead power supply. I was able to boot into LOV8 in maintenance mode and edit the fstab file. LOV8 now works OK but I have to add scratch. I am out of fuel tonight so this will have to wait until I come to CMU again. Predrag On Fri, Dec 9, 2022 at 5:46 PM Predrag Punosevac wrote: > Dear Autonians, > > Lov8 is refusing to reboot due to the faulty HDD used for scratch > directories. I am at CMU right now but not sure if I will be able to fix it > tonight. > > LOV4 is down reported without electricity. LOV4 is one of the oldest > servers in the lab originally provisioned by Kyle Miller before I came to > the Lab. Best case scenario for LOV4 is a dead power supply which is 5 > minutes fix. If it is a dead motherboard the server will have to be trashed > as it is older than 10 years. > > Best, > Predrag > -------------- next part -------------- An HTML attachment was scrubbed... URL: From awd at cs.cmu.edu Wed Dec 14 11:03:23 2022 From: awd at cs.cmu.edu (Artur Dubrawski) Date: Wed, 14 Dec 2022 11:03:23 -0500 Subject: REMINDER: Brief meetings with Artur this week In-Reply-To: References:

Message-ID: On Tue, Dec 13, 2022 at 1:01 PM Artur Dubrawski wrote: > I've just published some openings for this week. > > This week turned out to be quite out of sync for me and so I can only open > time on Fri and Sat. > I hope it will work for most of us. Next week will be my chemo treatment > week and I can't predict my availability yet. > > I would really need to meet with all PhD and MS student researchers before > all the winter breaks. > > Thanks > Artur > > >> >>> >>> On Mon, Nov 7, 2022, 4:35 PM Artur Dubrawski wrote: >>> >>>> The short meeting slots for this week have just been opened. Please >>>> book one (or more) while they last. >>>> >>>> Spreadsheet for signups link and the zoom link have not changed, they >>>> are below for easy reference. >>>> >>>> Cheers >>>> Artur >>>> >>>> >>>>> >>>>>>>> >>>>>>>>> >>>>>>>>> https://docs.google.com/spreadsheets/d/1OpY1DSxG7LLsMRroocMFTgqndSMiRTYUhYEwYdXH7Wc/edit?pli=1#gid=0 >>>>>>>>> >>>>>>>>> We will be using the same zoom link as before: >>>>>>>>> >>>>>>>>> https://cmu.zoom.us/j/9672166543 >>>>>>>>> >>>>>>>>> PS Let me know if the available times do not work for you and we >>>>>>>>> will >>>>>>>>> look for alternatives. >>>>>>>>> >>>>>>>> -------------- next part -------------- An HTML attachment was scrubbed... URL: From awd at cs.cmu.edu Thu Dec 15 13:17:35 2022 From: awd at cs.cmu.edu (Artur Dubrawski) Date: Thu, 15 Dec 2022 13:17:35 -0500 Subject: Fwd: Thesis Proposal - Dec. 16, 2022 - Sebastian Caldas - Collaborative learning by leveraging siloed data In-Reply-To: References: Message-ID: Team, Please come and see Sebastian give an excellent talk about his proposed thesis work. It happens tomorrow at 10:30am, on zoom. Cheers Artur ---------- Forwarded message --------- From: Diane Stidle Date: Mon, Dec 5, 2022 at 2:33 PM Subject: Thesis Proposal - Dec. 16, 2022 - Sebastian Caldas - Collaborative learning by leveraging siloed data To: ml-seminar at cs.cmu.edu , , < martin.jaggi at epfl.ch> *Thesis Proposal* Date: December 16, 2022 Time: 10:30am (EST) Remote Only Speaker: Sebastian Caldas *Title: Collaborative learning by leveraging siloed data* Abstract: Data holders cannot always share the data that they own, which can ultimately limit the modeling capabilities of each holder. For example, a hospital may lack representative records to learn about a new or rare condition, or a single mobile device may not have enough input to train a useful language model about its user. In both of these cases, these siloed data holders would benefit from collaborating with others in order to leverage their data. In recent years, the field of federated learning has taken an interest in learning performant collaborative models from siloed data. For these models to be truly useful, however, they must provide utility along dimensions beyond predictive performance, such as confidentiality, fairness and privacy. In this thesis proposal, I will demonstrate how to improve the utility of collaborative models that leverage siloed data, focusing on three dimensions of utility that are of current relevance to collaborative contexts: Explanations: We combine explanations with predictive performance in pursuit of true clinical utility. To this end, we introduced FRCLS, an algorithm that can explicitly identify when a prediction is using knowledge from an external collaborator, and provides interpretable rules that delineate subpopulations for which that external knowledge is useful. We have demonstrated the efficacy of FRCLS on a variety of clinical tasks including early prediction of sepsis and prediction of overly long lengths of stay. Expert supervision: We encode domain knowledge into on-device data, enabling collaborative learning for a wider variety of problems. We encode this knowledge by leveraging heuristics curated by experts. We first learn which heuristics will be useful for the devices? data and then train a weakly supervised federated model using these heuristics. Communication constraints: To complete my dissertation, I propose to study settings where collaborators are limited in the number of rounds of communication that can be exchanged, as is seen in clinical settings with limited infrastructure. I propose to develop an adaptive knowledge distillation strategy and to demonstrate it in a healthcare application context. *Thesis Committee:* Artur Dubrawski (chair) Virginia Smith Gilles Clermont (University of Pittsburgh) Martin Jaggi (EPFL) *Zoom meeting link:* https://cmu.zoom.us/j/94957077221?pwd=SkdHZitvNkR2Zm9lSXMyUGtPUldjQT09 *Link to the draft document: * https://drive.google.com/file/d/1Wu_ysaVm22G5PgOpTJbHy14DJ-uvr34U/view?usp=sharing -------------- next part -------------- An HTML attachment was scrubbed... URL: From predragp at andrew.cmu.edu Sat Dec 17 09:04:58 2022 From: predragp at andrew.cmu.edu (Predrag Punosevac) Date: Sat, 17 Dec 2022 09:04:58 -0500 Subject: bash is down Message-ID: I used late last night but it appears to be down. I taught it was power outage but the other machines on the same location can be pinged. I have no idea untill I have the physical access. -------------- next part -------------- An HTML attachment was scrubbed... URL: