Reminder this is in 8 minutes
From: Pl-seminar <pl-seminar-bounces@xxxxxxxxxxx> on behalf of JOHN CYPHERT <pl-seminar-bounces@xxxxxxxxxxx>
Sent: Thursday, October 15, 2020 1:52 PM
To: pl-seminar@xxxxxxxxxxx <pl-seminar@xxxxxxxxxxx>
Subject: [madPL] PL Seminar Tomorrow
Hi everyone,
We will be having a PL seminar tomorrow Oct
16 at 1pm.
Yuhao will be doing a two-part presentation. In the first half of the seminar Yuhao will present the ESEC/FSE20 paper Detecting
Numerical Bugs in Neural Network Architectures. In
the second half of the talk Yuhao will talk about some of his current research.
The abstract for the paper:
Detecting bugs in deep learning software at the architecture level provides additional benefits that detecting bugs at the model
level does not provide. This paper makes the first attempt to conduct static analysis for detecting numerical bugs at the architecture level. We propose a static analysis approach for detecting numerical bugs in neural architectures based on abstract interpretation.
Our approach mainly comprises two kinds of abstraction techniques, i.e., one for tensors and one for numerical values. Moreover, to scale up while maintaining adequate detection precision, we propose two abstraction techniques: tensor partitioning and (elementwise)
affine relation analysis to abstract tensors and numerical values, respectively. We realize the combination scheme of tensor partitioning and affine relation analysis (together with interval analysis) as DEBAR, and evaluate it on two datasets: neural architectures
with known bugs (collected from existing studies) and real-world neural architectures. The evaluation results show that DEBAR outperforms other tensor and numerical abstraction techniques on accuracy without losing scalability. DEBAR successfully detects all
known numerical bugs with no false positives within 1.7–2.3 seconds per architecture. On the real-world architectures, DEBAR reports 529 warnings within 2.6–135.4 seconds per architecture, where 299 warnings are true positives.
The informal abstract for Yuhao's current research project:
We study the robustness of NLP models over the discrete perturbation space that mimics spelling mistakes and other meaning-preserving transformations. We present
a language that allows the user to specify string transformations programmatically. We then show how to abstract the programmable perturbation space precisely for Recursive Neural Networks.
See you tomorrow,
John
Join Zoom Meeting
https://uwmadison.zoom.us/j/97079120314?pwd=UFN3c3lDcWlieWdqdjVUNnk4aVZXUT09
Meeting ID: 970 7912 0314
Passcode: 790380
One tap mobile
+19292056099,,97079120314#,,,,,,0#,,790380# US (New York)
+13017158592,,97079120314#,,,,,,0#,,790380# US (Germantown)
Dial by your location
+1 929 205 6099 US (New York)
+1 301 715 8592 US (Germantown)
+1 312 626 6799 US (Chicago)
+1 669 900 6833 US (San Jose)
+1 253 215 8782 US (Tacoma)
+1 346 248 7799 US (Houston)
Meeting ID: 970 7912 0314
Passcode: 790380
Find your local number: https://uwmadison.zoom.us/u/aiMtYuleM
Join by SIP
97079120314@xxxxxxxxxxx
Join by H.323
162.255.37.11 (US West)
162.255.36.11 (US East)
115.114.131.7 (India Mumbai)
115.114.115.7 (India Hyderabad)
213.19.144.110 (Amsterdam Netherlands)
213.244.140.110 (Germany)
103.122.166.55 (Australia)
149.137.40.110 (Singapore)
64.211.144.160 (Brazil)
69.174.57.160 (Canada)
207.226.132.110 (Japan)
Meeting ID: 970 7912 0314
Passcode: 790380
|