HNHacker News
TopNewBestAskShowJobs

aschampion

804 karma · joined July 10, 2013

submissionscomments
aschampion··on TensorFlow 1.0 Released
The URL when I first clicked this was the GitHub release notes, which is far more informative and apropos to the HN audience than either the TF landing page or vague VentureBeat pseudonews.
aschampion··on Modern Software Over-Engineering Mistakes
See one of many previous discussions here:

https://news.ycombinator.com/item?id=12718270

aschampion··on Georgia Tech Offers Online Master of Science in Analytics Degree for Under $10K
I agree that latency and detail of feedback is an enormous problem with this sort of partially-guided coursework. However, it's a generalized problem with higher education, not specific to GT, in that when implemented effectively it's one of the most valuable education experiences but difficult to scale, because it demands time-consuming supervision.

This is especially true of term project courses, where the final portion of the project to which you devote the most time and creativity is also the part for which you're likely to receive the least feedback.

aschampion··on Georgia Tech Offers Online Master of Science in Analytics Degree for Under $10K
> I did both my BSCS and MSCS at Georgia Tech. While I have many complaints about the school, the quality of the classes is not one of them, for either the undergrad or grad programs.

I also did BS and MS at GT, and while I generally share your experiences there were 3 or 4 truly disappointing classes during my MS. They didn't ruin my overall experience, but I can see how someone could happen to have more experiences like those and fewer positive ones and come aware with a very different perception of course quality.

My overall opinion of GT is mixed, but rigor or the courses is not one of my top critiques.

aschampion··on Georgia Tech Offers Online Master of Science in Analytics Degree for Under $10K
I did a resident MS CSE around the time they started the online program. While for the classes that were also available online they seemed equivalent (usually just recorded versions of the same class), there is one big caveat: most of the courses I took during my MS, or at least most of the most valuable courses I took, were not offered online.

My BS CS was also from GT (several years before), so much of the core course list available online were slightly more rigorous retreads of distributed algorithms, simulation architectures, etc. already covered late in undergrad. The most valuable courses in my MS were special topics courses that, at least at the time, weren't available online, many of which were in other departments as part of the computation + application domain interdisciplinary approach of the program.

That said, the core courses that were also offered online were generally quite good, and a few were exceptionally great. Rich Vuduc's HPC applications course comes to mind.

aschampion··on Welcome Chris Lattner
Language shapes perception shapes language, so it's absolutely worthwhile to be vigilant and critical about the idioms we endorse by adopting, especially for engineered phrases (e.g., "right to work").
aschampion··on Rust is more than safety
That's an odd example, given most video games are written in memory managed languages for performance reasons. Typically an allocation pool would be used for that case because there's some bound to the number of simultaneously allocated objects and each object has a homogeneous memory footprint.

It's also odd because you're replying to a well-known game designer.

That said, there are allocation-heavy workloads for which GC is well suited.

aschampion··on Rust is more than safety
A big non-safety selling point of Rust for me is a systems language with the tooling and ecosystem of a modern dynamic language. Dependency management and composition is a necessary part of complex programs, but a tedious, painful timesink in almost all systems languages. While CMake and similar tools improved the situation for C++, it's still an enormous pain and a barrier to sharing and reusing code. You are often forced to re-invent the wheel -- complete with your own bugs and maintenance obligation -- simply because getting a dependency to be built and detected cross-platform may be an even larger cost.

Cargo is a wonder. It's more limited in scope than something like CMake, but for the vast majority of use cases that are just pulling in a library or binding from the same language, it's spectacular.

Related to this, having an opinionated lint built into the tooling creates a common, readable dialect for the ecosystem. This is especially important for verbose syntax languages like Rust or C++. There are C++ libraries I can't understand simply because special snowflake formatting combined with complex template syntax renders it illegible.

aschampion··on Oculus CEO Brendan Iribe Steps Down
I use Facebook and owned a DK1 and DK2, but did not get the CV1 or continue to the use my DK2 because of the Facebook integration.

I trust Facebook with the information I explicitly share with it, i.e., posting photos or comments on Facebook. I do not trust Facebook to have unfettered access to a microphone, camera, and activity just to be able to play a game. The Facebook integration was, well, integral, and not just an opt-in for enhancing social aspects. You can't use the headset to play local content without phoning home (short of using old drivers/OpenVR or something).

Of course maybe this still places me in a minority of users because I have the same reaction when a mobile app requests every permission possible, which seems to be the case with all apps now.

aschampion··on MacBook Pro
Not wanting to divert attention from the screen during engaged work has nothing to do with fear of change.
aschampion··on Smartwatch sales are tanking
I like my 2nd gen Moto 360 as a watch, for its aesthetics, and as a simple notifier, but almost everything else on it is beyond broken. Just trying to check the weather takes 20+ seconds and fails half the time, even on wifi. Pull out the phone and the weather is already there, updated. What the hell. Ok Google could be useful, except for the situations you'd want to use it (hands full, crowded environment) are exactly the situations where it fails.

USAA has a great watch app, though, which somehow has many fewer reliability/latency problems than every other app. I can do most of my banking just walking down the hall while my phone is locked up at my desk.

aschampion··on The Myth of RAM (2014)
A complexity measure is a measure of complexity over a model. Mostly we talk about input size models, but we can (and do) talk about other models. It isn't a question of complexity scaling vs. time scaling, but of complexity scaling over an input size model vs. a machine architecture model.

There are tons of well-researched machine architectural models for complexity analysis. In the case of the article, any one of the many external memory (EM) or hierarchical memory models would be appropriate. These can capture latencies and bandwidth of hierarchical memories and storage, and in some cases even eviction, contention, partition, etc.

This isn't some esoteric point. In high performance computing, distributed algorithms, or even systems & architecture it's pretty much expected that if you talk about complexity, you should do so under some sort of EM model.

aschampion··on Does a compiler use all x86 instructions? (2010)
> If there is anything the article of this thread suggests, it is that we have created a needlessly complex instruction set, and that "efficient specialization" is not valuable to the software 99.99% of the time.

The only thing the article suggests is that the author's /usr/bin only includes about 2/3rds of possible x86 instructions, and that a few instructions occur most. The latter is unsurprising and expected in almost any grammar or formal system. The former is also expected given that most specialized instructions are for specialized use in high-performance applications, which are not likely all to be in any individual workstation's /usr/bin.

The cruft of specialization is legacy instructions, which as many other comments in this thread point out are usually implemented through microcode only. They don't contribute to complexity of the processor (which can choose to inefficiently implement this with other microinstructions) or the compiler (which can choose to not emit them).

I'm all for new blood in general-purpose architectures, but the bloat and inelegance of x86 isn't really something that's mattered for at least a decade.

aschampion··on Does a compiler use all x86 instructions? (2010)
We "over-complicate" ISAs for the same reason we're constantly adopting new vocabulary: there is efficiency in specialization. Good design is not about simplicity; it's about managed complexity.

> - Why do compilers not strive to simplify their code-gen phase, or enable themselves to do advanced instruction-level program analysis, or both?

Because specialized instructions formalize invariants and constraints on behavior that allow efficient computation, often by specialized hardware.

> - Why do microprocessors not strive for simplicity, implement only a handful of instructions in an optimized way, with a very small chip footprint, to be followed by proliferation of cores (think 256-core, 512-core, 1024-core).

Some do, see GPUs and coprocessors like the Phi. We don't take this approach with CPUs because real problems often require complex, branching, inhomogeneous computation, which require the type of specialization and tradeoffs mentioned above.

aschampion··on 455 Tb of live streaming were transferred during the Olympic Games Rio 2016
Watching via the BBC is a completely different experience: complete on-demand archives almost every event, time bookmarks for groups/bouts/matches, no annoying commentators unless the event was also being broadcast on a channel, no missing most of the event because we need to watch a manufactured interest piece on the US athlete, and you can go 30 seconds without hearing the name Phelps.

Watching NBC's broadcast, in contrast, made it seem like NBC had no belief that any of the events were interesting in and of themselves.

aschampion··on Computers do better than pathologists in predicting lung cancer type, severity
I worked on a similar pipeline for renal cell carcinoma a few years ago [1], although we only published a small subset of results since parts of the pipeline (e.g., finding representative tiles, survival prediction) had better results being produced elsewhere in the lab.

Regarding the hook in the headline -- computers surpassing pathologists -- it's a bit like automated driving in that even if true the immediate problem is the social and economic system. That is, we're not going to be removing the pathologist from the diagnostic and prognostic process anytime soon for many reasons, so how instead do we leverage machine learning in concert with the human observer to improve the diagnostic system? For that reason, decision introspection may be as valuable a topic of research as improving classification accuracy: justifying a particular automated classification to the pathologist, directing them to representative regions, and describing regions of feature space in biological terms.

[1] http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6945104

aschampion··on NASA plans to hand over the ISS to a private corporation
The reboosts are at hundredths of a g.
aschampion··on NASA plans to hand over the ISS to a private corporation
Russia has been planning (or at least posturing) to split off their modules in 2020 anyway.
aschampion··on NASA plans to hand over the ISS to a private corporation
AFAIK the ISS isn't very rigid and isn't designed to withstand high-g accelerations, hence the robotic arm docking, etc.
aschampion··on NASA plans to hand over the ISS to a private corporation
If someone was willing to take on the operating costs and still lease out access/experiment time for anything less than billions, NASA would be champing at the bit to give it away.

However, I can't imagine why any commercial entity would be interested in burning cash in LEO, unless they have a very lucrative microgravity production requirement. But so far that scenario has just been a fable.

aschampion··on Depth Precision Visualized
Coincidentally I was reading another good post on this same topic earlier today:

http://outerra.blogspot.com/2012/11/maximizing-depth-buffer-...

aschampion··on Master Plan, Part Deux
It's not a question of intelligence, but of attention. Few crashes are caused by wrong inferences; most are caused by lapses in attention or impaired judgement. Computer systems do not necessarily suffer from those.
aschampion··on Scaling nodes in a Dgraph cluster
Yours is a great example of what graph DBs should be good at, but many self-styled graph DBs out there at the moment are not. Graph DB means to me only two things: index-free edge traversal and scalable built-in graph operations. While these would seem to be necessary and sufficient criteria to distinguish a graph DB, some instead use only the criteria of the GGP and equivocate graph DBs with schema-free DBs, which should be orthogonal axes of database features.
aschampion··on Valve Steam Machine selling very poorly
While I don't think Steam machines are DOA quite yet, they target an already difficult niche for which Valve's indecisive approach is ill-suited.

The lack of marketing and immaturity of platform and user experience mean the most viable demographic is existing PC gamers. However, existing PC gamers are adverse to markups for what are basically SFF PCs, and are much more inclined to build their own (or buy a lower markup, non-Steam branded PC) to fill that niche. More affluent parts of the demographic less concerned about markup also tend to be older and prefer something more subtle than the gaming glitz styling and branding of most of the premium Steam machines.

There's also self-cannibalism in this demo. Few of the Steam machines are top-tier components for the sake of heat and noise, so it's easy for most existing PC owners to fill this niche by simply replacing their current PC (as most do every few years) and demoting the old box to the TV room. As you mentioned, Steam link or NVIDIA's in-home streaming on Shield TV can also fill most use cases (at least they do for me).

Valve could have leveraged the initial hype and novelty of Big Picture and the Steam machine concept to drive an initial wave of sales, but the delay between those announcements and a consequential number of actual machines being ready to buy outlasted the initial hype.

aschampion··on CryEngine out on GitHub
I haven't look through the repo yet, but I know there was some shake-up a while back with the Star Citizen developers injecting capital into CryEngine. I don't care about the politics, but I am curious how the branch published here relates to branches used by developers on those projects, i.e., will things like their 64-bit fork be merged at some point?
aschampion··on A Reboot of the Legendary Physics Site ArXiv Could Shape Open Science
My favorite is pop science reporting never linking to the press release for the publication that it's pulling from, much less to the publication itself. Often the title of the article isn't even mentioned.
aschampion··on Performance of ES6 features relative to ES5
This is the convention when talking about speedup in software. 1x speedup means identical time.

[1] https://en.wikipedia.org/wiki/Speedup

aschampion··on OpenSSL CVE-2016-0799: heap corruption via BIO_printf
> These things can be reduced to nuisance level by discipline.

Obviously not. We've been doing this for decades and are still making the same mistakes with costly effects. Humans are as much a part of technological systems as the language or platform; continuing to ignore that even educated, disciplined, skilled developers are prone to certain classes of errors -- errors that could be obviated by other components of the system -- is hubris. It is much more feasible to create tools that exclude certain classes of error than to create less fallible developers.

> (code size; execution speed; random latencies)

Rust is not GC'ed, has no inherent reason to be slower than C (as the toolchain matures), and I doubt is more verbose than C code that provides the same safety.

aschampion··on Records: Python library for making raw SQL queries to Postgres databases
Good point -- I was talking in the context of bulk upserts but didn't make that explicit.
aschampion··on Records: Python library for making raw SQL queries to Postgres databases
Neat library. Just to note that this approach to fake upsert is not safe to concurrent sessions. There were exactly two concurrency-safe ways to implement upsert prior to 9.5:

1. An exclusive lock on the table, which is slow as hell 2. A PGPSQL function that attempts to INSERT, catches the duplicate key exception and instead UPDATEs (also handling the case that the row vanishes), which is slower than hell

This is why ON CONFLICT is such a huge deal. If it were as simple as checking keys with a SELECT then sending some to INSERT and some to UPDATE it would just be nice syntax sugar.

← PreviousPage 3 of 5Next →