I give many talks and since there is a significant overlap among them, it does little good to have copies of each set of slides available. Instead, this page attempts to give a thinned sample of slides from talks that are coarsely broken out by topic area. There is still some overlap, but there should be less. If you are looking for the slides corresponding to a talk that you've heard me give, then it should be easy to find the 1-3 slide-sets that cover the same material in roughly the same (or an improved) manner. For the most part, I have included the most recent material on the topic. So older talks will have pointers to updated slides.
Of course, all of these slides were meant to complement my talk, not to be a standalone substitute for the talk itself. Ideally, I'd have recordings or something to go along here, but you're out of luck since I have not recorded my talks.
The papers corresponding to the materials in the talk can be found here for cognitive radio related work and here for more information-theoretic work. Feel free to email me if you have questions, but I apologize in advance if I do not have the time to answer your emails. Ideally, I would put more specific pointers here itself, but I haven't gotten around to doing that yet. Finally, in Fall 2008, I was inspired by reading the work of Edward Tufte and moved my group to a new model of slides (with essentially no bullets) plus handouts distributed to the audience. This newer model seems to work much better, but the older material below has not migrated to the new style.
|Topic||Talk Date and Location||Talk title and brief comments|
|Cognitive Radio||Y 2/09||Seminar at Harvard||"Prospects and challenges for spectrum sharing by cognitive radios" (slides)||This post-FCC ruling talk uses the DTV bands to illustrate how much white-space is out there and the core underlying tradeoffs involved, including the intrinsically political perspective of trading off users of two different possible services: television vs white-space devices. The talk introduces the F_HI metric for safety and shows how 1-WPAR is a reasonable metric for the regulatory overhead imposed by a sensing strategy. Because single-user sensing has terrible overheads, the question becomes how to enable multiuser sensing --- and in particular how to regulate it. Our new "criminal law" perspective on light-handed regulation is introduced and the required overhead of robust identity and reliable deterrence is explored.|
|A 6/08||Keynote at IEEE Workshop on Networking for SDR in San Francisco||"New directions in cognitive radio and spectrum sharing"||This pre-FCC ruling talk also gives a good overview and introduces the problem of enforcement as well as the need for our new metrics of F_HI and WPAR.|
|B 1/08||Invited Lecture at the Institute for Infocomm Research in Singapore||"SNR Walls for Cognitive Radio Spectrum Sensing"||This dwells on the impact of uncertainty in cognitive radio. Beyond the main material on SNR Walls and the causes behind fundamental limits to spectrum sensing, there is some material up front on why PHY-layer approaches to coexistence using Dirty-Paper-Coding are hard to make work in practice with primary users.|
|C 11/05||Tutorial at the first DySpAN conference in Baltimore||"Spectrum Sensing: fundamental limits and practical challenges: Part I and Part II"||This introduced many important ideas to the cognitive radio community in a unified way. The power-control ideas were particularly important and not covered in the talks above.|
|Complexity and Power Consumption||D 4/08||Seminar at MIT in Cambridge||"Fundamental bounds for power consumption at the physical layer: `waterslide curves' and the price of certainty" (related handout)||This talk reviews our approach to power consumption, shows why classical codes are bad, and illustrates the fundamental limits on total power consumption given our idealized model of computation.|
|Mixed Topics||Z 5/09||Seminar at UIUC||"What does dirty paper coding have to do with distributed control? Revisiting the Witsenhausen counterexample" (slides only)||This unveiled our new perspective on the 40-year-old Witsenhausen counterexample: a problem that captures the essence of the implicit interactions that are possible in distributed control systems. We take an information-theoretic view inspired by the "cognitive radio channel" and deterministic channel models that leads directly to an asymptotically infinite-dimensional version of the counterexample for which new upper and lower bounds reveal that it is always possible to achieve within a factor of two of the optimal cost. The results are then pulled back to the finite dimensionsonal case using our "Platonic perspective" on sphere-packing bounds to get a similar factor of eight approximate optimality result for Witsenhausen's original scalar counterexample.|
|E 3/08||Seminar at the Global COE Workshop on Networked Control Systems at Kyoto University||"The connection between information theory and networked control"||This talk emphasizes the connections between control and information theory. It starts with our results on stabilization over noisy channels, goes through anytime bounds with/without feedback, reinterprets them back in the control context, and closes with a discussion of how our work on power consumption and complexity is really about viewing decoding as a distributed control problem.|
|F 3/07||Presentation at the Conference for Information Sciences and Systems in Baltimore, MD||"Universal Quadratic Lower Bounds on Source Coding Error Exponents"||This talk focuses on how to get universal error exponents. The main point of interest is a correction to the solution of a problem of Gallager that is then generalized to cover source coding. This is then also evaluated for streaming problems.|
|G 4/07||Presentation at the Workshop on Control over Communication Channels in Limassol, Cyprus||"Universal Anytime Codes: An approach to uncertain channels in control"||This talk uses the universal error exponents in the control context. The main point of interest is how to generalize universal decoding into the streaming context where there is no blocklength.|
|Delay, Anytime codes, Noisy Feedback||H 6/07||Tutorial at ISIT in Nice, France||"Feedback and Side-Information in Information Theory"||This gives a broad overview of many of my results in this area including the error probability as a function of delay for both "hard" and "soft" deadline problems as well as the distinction between block and nonblock problems. The role of noisy/unreliable feedback is also touched upon.|
|I 3/08||Seminar at the University of Tokyo||"How valuable is noisy or unreliable feedback"||This takes a closer look at the problem of improving reliability using only noisy and unreliable feedback. This looks at the Burnashev-style variable-length block setting, the Kudryashov setting of variable-delay streaming, and the fixed-delay streaming setting. There is no reliability improvement for fixed-length blocks so that setting is moot.|
|J 7/08||Talk at ISIT 2008||"The `hallucination' bound for the BSC"||This gives more details of the proof and motivation behind the Hallucination Bound for the fixed-delay soft deadline setting with feedback.|
|K 7/06||Lecture at the Summer Research Institute in the Ecole Polytechnique Federale de Lausanne, Switzerland||"Delay, feedback, and the price of ignorance"||This focuses on the problem of fixed-delay streaming communication. There is substantial overlap with the tutorial above, but there are some extra slides that explain more details of the proofs.|
|L 7/08||Talk at ISIT 2008||"Trade-off of lossless source coding error exponents"||This gives some new inner and outer bounds for the reliability/rate regions for multi-stream source coding. The talk also includes the erasure channel case of anytime channel coding with feedback that is particularly relevant to control settings.|
|M 9/05||Presentation at ISIT in Adelaide, Australia||"Anytime communication over the Gilbert-Eliot channel with noiseless feedback"||This is an example of a talk wherein we show how to compute the anytime capacity with feedback. There were talks that we gave at Allerton earlier, but this one is notable for my introduction of this deterministic "escalator" imagery for how reliable communication proceeds in the S-K scheme for Gaussian channels by lifting bits steadily above the noise floor. The Gilbert-Eliot channel then had a semi-deterministic picture for it. This sort of imagery then went on to inspire one of these students to do some great work with David Tse on deterministic channel models for interference networks. The other went on to do nice work with Massimo on control problems and interference.|
|N 11/04||Presentation at ITW in San Antonio, TX||"On the variable-delay reliability function of discrete memoryless channels with access to noisy feedback"||This is the first talk I gave on the issue of noisy feedback in the Burnashev setting. Technically, it is almost entirely superceded by later talks, but it is mildly interesting from a historical perspective.|
|Control with Communication Constraints||O 7/06||Presentation at the International Symposium on Mathematical Theory of Networks and Systems in Kyoto, Japan||"Stabilization using both noisy and noiseless feedback"||This covers our basic results on control systems, including how we can use sequential decoding ideas to actually give a semi-implementable nonlinear control strategy. This also covers our interpretation of the "fortified" bounds for anytime codes in terms of a combination of noiseless and noisy feedback for control systems with dynamic state sampling.|
|P 4/05||Seminar at the Caltech Workshop on Integrated Control, Estimation and Communication in Pasadena, CA||"Where do stabilization problems sit in the communication problem hierarchy?"||This gives our interpretation of our control results in terms of a hierarchy of communication problems. We later developed this insight as we revisited the classical source/channel separation theorem.|
|Q 12/04||Presentation at the Conference on Decision and Control in Paradise Island, Bahamas||"The necessity and sufficiency of anytime capacity for control over a noisy communication link"||This gives our basic scalar results for control systems. The talk is useful because it has slides covering the sufficiency aspects of the problem.|
|R 8/04||Seminar at the University of Illinois at Urbana Champaign||"Control, feedback, and the nature of information"||This talk reviews the basic results for stabilization (significant overlap with the CDC talk above) and estimation of unstable processes and also reveals how we started zooming into the problem of fixed-delay communication. This was right before we realized that Pinsker's result was wrong.|
|Source/Channel Separation||S 11/06||Lecture for our advanced information theory course||"An AVC perspective on source/channel separation"||This gives a new perspective on the classical source/channel separation theorem by establishing an operational equivalence between the lossy communication of information and traditional reliable communication. In a sense, this can be interpreted as the fundamental theorem of steganography. These ideas are very useful for proving more nuanced converse bounds when the information structure is not purely classical. [Some images taken from Babylon 5 and the movie Bedazzled.]|
If I have seen or heard me give a talk at your institution whose content is not covered by the above sets of slides, please let me know and I will attempt to post the relevant slides online for your convenience.
Special thanks to our past and present research sponsors: