From siddiqi at cs.cmu.edu Fri Oct 9 15:20:07 2009 From: siddiqi at cs.cmu.edu (Sajid Siddiqi) Date: Fri, 9 Oct 2009 15:20:07 -0400 (EDT) Subject: [auton-users] Fwd: Thesis Oral: Sajid M. Siddiqi Message-ID: <1739.128.2.182.68.1255116007.squirrel@webmail.cs.cmu.edu> My thesis defense is on Thursday at noon in the room around the corner. Come cheer me on / throw rotten eggs, as required :) I hear that there will be food. Sajid --------------------------------------- Date: 15 October 2009 Time: 12:00 p.m. Place: Newell Simon Hall 3305 Type: Thesis Oral Who: Sajid M. Siddiqi Topic: Learning Latent Variable and Predictive Models of Dynamical Systems Abstract: A variety of learning problems in robotics, computer vision and other areas of artificial intelligence can be construed as problems of learning statistical models for dynamical systems from sequential observations. Good dynamical system models allow us to represent and predict observations in these systems, which in turn enables applications such as classification, planning, control, simulation, anomaly detection and forecasting. One class of dynamical system models assumes the existence of an underlying hidden random variable that evolves over time and emits the observations we see. Past observations are summarized into the belief distribution over this random variable, which represents the state of the system. This assumption leads to `latent variable models' which are used heavily in practice. However, learning algorithms for these models still face a variety of issues such as model selection, local optima and instability. The representational ability of these models also differs significantly based on whether the underlying latent variable is assumed to be discrete as in Hidden Markov Models (HMMs), or real-valued as in Linear Dynamical Systems (LDSs). Another recently introduced class of models represents state as a set of predictions about future observations rather than as a latent variable summarizing the past. These `predictive models', such as Predictive State Representations (PSRs), are provably more powerful than latent variable models and hold the promise of allowing more accurate, efficient learning algorithms since no hidden quantities are involved. However, this promise has not been realized. In this thesis we propose novel learning algorithms that address the issues of model selection, local minima and instability in learning latent variable models. We show that certain 'predictive' latent variable model learning methods bridge the gap between latent variable and predictive models. We also propose a novel latent variable model, the Reduced-Rank HMM (RR-HMM), that combines desirable properties of discrete and real-valued latent-variable models. We show that reparameterizing the class of RR-HMMs yields a subset of PSRs, and propose an asymptotically unbiased predictive learning algorithm for RR-HMMs and PSRs along with finite-sample error bounds for the RR-HMM case. In terms of efficiency and accuracy, our methods outperform alternatives on dynamic texture videos, mobile robot visual sensing data, and other domains. Thesis Committee Members: Geoffrey J. Gordon, Chair Andrew Moore Jeff Schneider Zoubin Ghahramani, University of Cambridge David Wingate, Massachusetts Institute of Technology A copy of the thesis document is available at: http://www.cs.cmu.edu/~siddiqi/docs/sajid_thesis_draft.pdf From mjbaysek at cs.cmu.edu Tue Oct 20 10:21:44 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Tue, 20 Oct 2009 10:21:44 -0400 Subject: [auton-users] Compute Machine Refresh: Response Requested Message-ID: <4ADDC778.6030400@cs.cmu.edu> Hi Lab. I notice that overall, the compute machines have not been very busy recently. I want to take advantage of this, and take time very soon (within the next week or two) to upgrade all of the compute machines to newer OS and software versions. If nobody replies with any mission critical jobs they are going to be running, I will proceed to take down probably all of the lop/q/r/s's (except one) and upgrade them all at once. This will require a full work day of down time for the affected machines. Please let me know if you have any plans to run any time critical or long running experiments coming up, so I can plan accordingly. Mike -- -- Michael J. Baysek, Systems Analyst Carnegie Mellon University - Auton Lab www.cmu.edu - www.autonlab.org 412-268-8939 From mjbaysek at cs.cmu.edu Tue Oct 20 15:34:20 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Tue, 20 Oct 2009 15:34:20 -0400 Subject: [auton-users] Compute Machine Refresh Update In-Reply-To: <4ADDC778.6030400@cs.cmu.edu> References: <4ADDC778.6030400@cs.cmu.edu> Message-ID: <4ADE10BC.1050701@cs.cmu.edu> Given the responses I have received so far, I would like to do this update on Friday this week. This seems to impact everyone the least. Let me know if there is anyone with a critical reason Friday will not work. -Mike -- Michael J. Baysek, Systems Analyst Carnegie Mellon University - Auton Lab www.cmu.edu - www.autonlab.org 412-268-8939 Michael J. Baysek wrote, On 10/20/2009 10:21 AM: > Hi Lab. > > I notice that overall, the compute machines have not been very busy > recently. I want to take advantage of this, and take time very soon > (within the next week or two) to upgrade all of the compute machines > to newer OS and software versions. > > If nobody replies with any mission critical jobs they are going to be > running, I will proceed to take down probably all of the lop/q/r/s's > (except one) and upgrade them all at once. This will require a full > work day of down time for the affected machines. > > > Please let me know if you have any plans to run any time critical or > long running experiments coming up, so I can plan accordingly. > > > Mike > > From mjbaysek at cs.cmu.edu Thu Oct 22 08:33:59 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Thu, 22 Oct 2009 08:33:59 -0400 Subject: [auton-users] Reminder Compute Machine Downtime Message-ID: <4AE05137.4090407@cs.cmu.edu> Hi Lab, This is a reminder that the compute machines will be down tomorrow Friday for a software upgrade. I am budgeting the entire day for the process, though it is quite likely many machines will be back online by around noontime. If you are planning jobs for after the downtime, please email me with your memory and CPU requirements and I can notify you as soon as a machine is online that matches your criteria. Mike From mjbaysek at cs.cmu.edu Fri Oct 23 15:33:00 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Fri, 23 Oct 2009 15:33:00 -0400 Subject: [auton-users] Lab Machine Upgrades Message-ID: <4AE204EC.8020000@cs.cmu.edu> Hi Lab, This a brief email because I am still working on this, but the lab machines are, for the most part back online. There are some software packages missing, notably VNC, MATLAB and JAVA. I will be adding these shortly. For VNC users, for now, use screen. If you don't know how, login to https://www.autonlab.org/auton_intranet and search for screen. There are three very helpful documents with varying levels of detail. Also, the following T-Cube Web Interfaces are down (but next to be restored): LOP6 (Sri Lanka) and LOS1 (Public Demo). USAF and USDA interfaces are UP. I will keep you informed of changes, and let you all know when I am done. LOQ2 and LOR1 will remain down over the weekend. LOQ1 and LOR2 however, have been unavailable, but are now UP, so it's kind of a wash. LINA and LIRA are not yet upgraded. These upgrades will happen early next week. The only people using them are Saswati, myself, and USDA T-Cube Web Interface users, but I thought it worth mentioning. Mike -- -- Michael J. Baysek, Systems Analyst Carnegie Mellon University - Auton Lab www.cmu.edu - www.autonlab.org 412-268-8939 From mjbaysek at cs.cmu.edu Fri Oct 23 17:04:46 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Fri, 23 Oct 2009 17:04:46 -0400 Subject: [auton-users] Lab Machine Upgrades In-Reply-To: <4AE204EC.8020000@cs.cmu.edu> References: <4AE204EC.8020000@cs.cmu.edu> Message-ID: <4AE21A6E.4040302@cs.cmu.edu> There is a new version of GCC on this install, so it will be very smart for you to clean all your object and binary files out of your h/ by running make zapsibs before recompiling, or even better, deleting all the ~/h/*Linux*gcc* directories. If you have any compilation problems, let me know. -- Michael J. Baysek, Systems Analyst Carnegie Mellon University - Auton Lab www.cmu.edu - www.autonlab.org 412-268-8939 Michael J. Baysek wrote, On 10/23/2009 03:33 PM: > Hi Lab, > > > This a brief email because I am still working on this, but the lab > machines are, for the most part back online. > > > There are some software packages missing, notably VNC, MATLAB and > JAVA. I will be adding these shortly. For VNC users, for now, use > screen. If you don't know how, login to > https://www.autonlab.org/auton_intranet and search for screen. There > are three very helpful documents with varying levels of detail. > > Also, the following T-Cube Web Interfaces are down (but next to be > restored): LOP6 (Sri Lanka) and LOS1 (Public Demo). USAF and USDA > interfaces are UP. > > > I will keep you informed of changes, and let you all know when I am done. > > > LOQ2 and LOR1 will remain down over the weekend. LOQ1 and LOR2 > however, have been unavailable, but are now UP, so it's kind of a wash. > > > LINA and LIRA are not yet upgraded. These upgrades will happen early > next week. The only people using them are Saswati, myself, and USDA > T-Cube Web Interface users, but I thought it worth mentioning. > > > Mike > From mjbaysek at cs.cmu.edu Fri Oct 23 23:59:30 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Fri, 23 Oct 2009 23:59:30 -0400 Subject: [auton-users] gmake-magic update required Message-ID: <4AE27BA2.5040703@cs.cmu.edu> Hi Lab, Two people have had problems building code which links to system includes on the new system. The fix has been committed to gmake-magic. You should update your ~/h/gmake-magic asap if you want to avoid this problem. As a side note: there have been a lot of fixes in the core code recently. You should all consider running cvs up on the whole of your ~/h directory by: cd ~/h cvs up As soon as you can. You should be in the habit of doing this at least monthly. We are soon releasing a system which will let you know about core code changes automatically, but for now, just keep updating your ~/h at sane intervals. Best, Mike From mjbaysek at cs.cmu.edu Sat Oct 24 17:13:22 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Sat, 24 Oct 2009 17:13:22 -0400 Subject: [auton-users] Matlab now available Message-ID: <4AE36DF2.8070508@cs.cmu.edu> Matlab is now available on the compute machines. Check the status page for available machines, since some machines are still down. -Mike From mjbaysek at cs.cmu.edu Thu Oct 29 13:42:25 2009 From: mjbaysek at cs.cmu.edu (Michael J. Baysek) Date: Thu, 29 Oct 2009 13:42:25 -0400 Subject: [auton-users] New Auton Compute Node: LOT2 Message-ID: <4AE9D401.5080108@cs.cmu.edu> A powerful new compute node is available. LOT2 does not have a public IP address, so to access it you must first ssh into LOP1. Number CPU Cores: 16 Processor cores System RAM: 64 GB RAM CPU GHz: 2.4 GHz CPU Type: AMD Opteron K10 Since this machine is such a big machine, it will be very likely to be in use by many people at once. By all means, get the most use out of this machine as you can! For example, RAM divided by CPU cores is 4 GB per CPU... If no jobs on the machine are using more than 4 GB, then you are welcome to launch jobs until the load average reaches 16. However, if any job (by any user) is consuming more than 4 GB of resident memory, you will want to be very careful about loading up all of the CPUs since you could run out of memory before you run out of CPUs. This new machine has already been added to the status page. When planning your jobs, if you are worried about resource contention with other users (especially when you're on a deadline) you can always send an email to me and I can help coordinate the resources. -- -- Michael J. Baysek, Systems Analyst Carnegie Mellon University - Auton Lab www.cmu.edu - www.autonlab.org 412-268-8939