US 9451883

I'll search for this specific patent number and any related litigation.

Let me run a few more targeted searches to confirm litigation status and claim details.

Let me verify litigation status and retrieve the full claim set.

US Patent 9,451,883 — Analyst Summary

Search scope and results

I ran targeted queries against the patent number, its kind code, and its digits (with/without commas) across Google Patents, USPTO-linked sources, PatentForecast, and general web/Federal Circuit docket-adjacent sources.

Litigation check — negative result. I found no 2026 CAFC docket, no PTAB proceeding, and no district court complaint asserting 9,451,883. Every litigation-flavored hit returned by the searches involved different patent numbers (e.g., 10,894,503; 9,917,563; 8,717,101; 6,890,188). I am stating this as an explicit negative finding rather than a confirmed "no litigation exists," because I could not query PACER, Docket Navigator, or the CAFC's own docket system directly — my tools returned indexed web content only. Treat "no litigation found" as absence of evidence, not evidence of absence.

Bibliographic data

Priority chain (from the specification): this application is a continuation-in-part of Ser. No. 12/715,557, filed 2010-03-02, which is a nonprovisional of provisional Ser. No. 61/157,310, filed 2009-03-04. Note the 2009-03-04 priority date and the 2012-12-21 CIP filing date are consistent with the record; the 2030-04-20 adjusted expiration reflects the pre-URAA-style term adjustment associated with the 2009 priority claim.

Abstract (verbatim substance)

Decoding and reconstructing a subjective perceptual or cognitive experience is described. A first set of brain activity data produced in response to a first brain activity stimulus is acquired from a subject using a brain imaging device. An encoding model is used to convert the brain activity data into a corresponding set of predicted response values. A second set of brain activity data produced in response to a second brain activity stimulus is acquired from a subject and decoded using a decoding distribution derived from the encoding model, and the probability the second set of brain activity data corresponds to said predicted response values is determined. The second set of brain activity stimuli is then reconstructed based on the probability of correspondence between the second set of brain activity data and the predicted response values.

Technology framing (brief)

The invention's stated point of novelty is a departure from conventional classification-based brain decoding (which can only identify stimuli from categories the classifier was trained on) toward reconstruction of arbitrary and novel perceptual/cognitive content. The mechanism is a linearizing feature space: the stimulus (or mental state) is projected nonlinearly into a feature space chosen so that the feature-to-brain-activity mapping is as linear as possible; separate encoding models are fit per measurement channel (voxel/sensor), then inverted into a decoding distribution. The specification is explicitly modality-agnostic as to both measurement technique (EEG, MEG, fMRI, fNIRS, SPECT, ECoG, "or any future brain measurement method") and brain system (vision, audition, touch, planning, imagery). Illustrated encodings include Gabor wavelet, scene category, motion-energy, and WordNet models for vision, and spectral/MPS, phonemic, syntactic, and semantic (LSA) models for audition/language.

Plain-language overview of the claims

Important caveat on completeness. The authoritative text I retrieved contains the specification and abstract, but the full verbatim granted claim set was truncated in the source feed. The overview below is compiled from (a) claim-language passages that appear in the retrieved text and (b) claim-text snippets surfaced via PatentForecast showing claims numbered at least into the 13–17 range. I cannot certify the exact claim count or the precise numbering of the independent claims. Flagging this rather than guessing.

Method claim — decoding via a decoding distribution (appears as claim 1)

1. Acquire a first set of brain activity data from a subject using a brain imaging device, where that data was produced in response to a first set of brain activity stimuli.
2. Convert that first dataset into a corresponding set of predicted response values.
3. Acquire a second set of brain activity data from the subject using the same imaging device, produced in response to a second set of brain activity stimuli.
4. Decode the second dataset using a decoding distribution, and determine a probability that the second dataset corresponds to the predicted response values.
5. Reconstruct the second set of stimuli based on that probability of correspondence.

Plain language: train a model on one brain scan session, then take a new scan and work backward to infer/reconstruct what the person saw, heard, or thought.

Method claim — encoding-model construction variant (independently recited)

1. Acquire first brain activity data produced in response to a first external stimulus, a first mental state, or a first cognitive state.
2. Construct an encoding model that predicts brain activity by projecting that stimulus/state nonlinearly into a feature space and finding weights that optimally predict brain activity from those features using regularized linear regression.
3. Acquire second brain activity data from the same subject or a different subject, produced in response to a second external stimulus / mental state / cognitive state.
4. Decode the second dataset (claim text truncated at this point in the source feed) — the balance of the claim is understood to complete the decode-and-reconstruct sequence, but I cannot quote it verbatim.

Plain language: the express-recitation claim, where the "linearizing feature space + regularized linear regression" architecture is a required claim element rather than merely descriptive.

Apparatus claims (two of them, mirroring the two method claims)

Both are in the form: a processor plus programming executable on the processor for performing the identical step sequences recited in the corresponding method claim. So the apparatus coverage is software-embodiment coverage of the same method.

Dependent claims — the notable ones

• Imaging modality: first and/or second brain activity data acquired using a technique selected from the group consisting of EEG, MEG, fMRI, fNIRS, SPECT and ECoG (a closed Markush group).
• Stimulus type: first and/or second stimuli selected from sensory stimuli, motor stimuli, and cognitive stimuli, or any combination thereof.
• Decoding-distribution mathematics: the distribution is positively recited as comprising a formula in which α = value of the second brain activity stimulus, r = measurement of the second brain activity dataset, s = the second set of stimuli, and h = adjustable parameters of an encoding model relating the first dataset to the first stimulus.
• Constructing the distribution: by evaluating p(r | s, h), then integrating over all possible parameter values with respect to p(h), and integrating over all possible stimuli with respect to p(s | α).
• Vision encodings: an encoding model selected from Gabor wavelet, scene category, motion energy, and WordNet.
• Auditory encodings: an encoding model selected from spectral, phonemic, syntactic, and semantic.
• Reconstruction by aggregation: per the PatentForecast snippet on the apparatus claim family, the reconstructed stimulus set is produced by computing an average or weighted average of selected one or more stimulus items — i.e., the decode step selects a top-N candidate set from a decoding database (a large corpus of same-type stimuli from the internet, a wearer's own continuously-recording camera, or a generative model), rather than requiring a single best match.

Uncertainty disclosures

• Litigation/CAFC: no 2026 (or other) CAFC docket entry, IPR, or district court case referencing 9,451,883 was found. This is a search-limited negative, not a definitive all-courts clearance.
• Claim set: the independent-claim overview is reconstructed from partially retrievable claim text. Exact independent-claim count, total claim count, and claim numbering are not verified against the granted patent's claims as issued. The snippets I did see (claims 13–17, an apparatus family, with claim 17 following) imply at least 17 claims and at least two apparatus independents, consistent with the four independent claims implied by the specification excerpts.
• Assignee: the "current assignee" listing (University of California / UC San Diego UCSD) and the recorded assignment (Regents of the University of California) are treated as consistent; no intervening assignment or expiry-of-rights event beyond "Expired – Fee Related" was surfaced.
• Status date: "Expired – Fee Related" with adjusted expiration 2030-04-20 was reported by the source; the source itself disclaims that legal status is a legal conclusion.

I'll research litigation involving US patent 9451883 across litigation-tracking sources.

Let me search litigation-specific databases directly.

Let me search more targeted litigation databases and free-form sources.

Summary of Findings

No litigation involving U.S. Patent No. 9,451,883 was surfaced by my searches. Based on the searches available to me, I could not identify any district court action, ITC investigation, PTAB proceeding, or Federal Circuit appeal in which U.S. Patent No. 9,451,883 ("Apparatus and method for decoding sensory and cognitive information from brain activity," assigned to the Regents of the University of California / UC San Diego) was asserted or challenged.

Important caveat — literal identifier matching

Per your instruction to interpret identifiers literally and not auto-correct, I want to be explicit about a significant source of noise I encountered: search results for the string "883" repeatedly returned different patents whose numbers merely contain the same terminal digits. None of these are U.S. 9,451,883, and none should be attributed to it:

I flag these because they are the kind of near-miss hits that a less careful search could wrongly report as "litigation involving 9451883." They are not.

Why absence here should be treated cautiously

• U.S. 9,451,883 issued in 2016 and, per Google Patents, is expired – fee related (adjusted expiration noted 2030-04-20), with the Regents of the University of California as assignee. A university-owned patent that has lapsed for failure to pay maintenance fees is a plausible candidate for never having been litigated, which is consistent with what I found.
• However, I was unable to run the full set of dedicated litigation tools (Unified Patents Portal case list, CourtListener/RECAP, Docket Alarm, PACER, and CAFC docket searches) to completion — my search budget was exhausted mid-query. Those are precisely the databases most likely to hold a definitive "no cases" or a low-visibility case.

Recommendation / status

Current status: No known litigation identified for U.S. 9,451,883, but this is not a certified negative.

To make this determination robust before you rely on it (e.g., for a freedom-to-operate or IP due-diligence opinion), I recommend the following confirmatory checks, which should ideally be run directly:

1. Unified Patents Portal litigation case list — search by patent number 9451883 (portal.unifiedpatents.com/litigation).
2. CourtListener / RECAP (free) and Docket Alarm or Lex Machina (paid) — query the patent number as a party field.
3. USPTO Patent Center / PTAB — check the patent's "Proceedings" and "Assignment" tabs for any IPR/PGR or recorded transfers indicating enforcement activity.
4. CAFC docket — check for appeals captioned with the patent number.
5. UC San Diego / UC Regents Office of Technology Transfer — university patent holders sometimes enforce through exclusive licensees, who may appear as the named plaintiff rather than the Regents.

If you'd like, I can attempt additional targeted searches (e.g., by assignee "Regents of the University of California" + brain decoding, or by known licensee names) once more search capacity is available.

I'll verify the structured data against public sources before reporting.

Proceedings overview

Total AIA trial proceedings on file for US 9,451,883: zero (0). The USPTO Open Data Portal returns no IPR, PGR, or CBM proceeding for this patent, and independent web searching (PTAB petition/Document repositories, PTAB decision coverage, and litigation-tracking hits) surfaced no petition, institution decision, or Final Written Decision naming the Gallant et al. patent. Breakdown by status is therefore 0 active / 0 claims invalidated / 0 claims sustained / 0 settled / 0 institution denied.

Bottom line for a defendant: no AIA trial has ever been filed against this patent, so there is no PTAB record to lean on — no canceled claims to cite, no FWD to quote, no § 315(e)(2) estoppel to invoke against anyone. The defensive posture is the unassisted one: you would be the first petitioner, and validity would have to be attacked from scratch. Two caveats temper that: (1) the patent's Google Patents legal status reads "Expired - Fee Related, expires 2030-04-20," which suggests a maintenance fee has lapsed (see next steps — verify on Patent Center before relying on it); and (2) an expired or unasserted university-owned patent of this age rarely justifies IPR spend, since the real defense is usually standing, ownership, or damages rather than patentability.

No proceeding to report — confirmation of the negative finding

There is no ### {PROCEEDING_NUMBER} — {Petitioner} v. {Patent Owner} entry to write, because no such proceeding exists. Rather than pad this section with invented numbers, here is what was checked and what was ruled out:

• Structured source (authoritative): the USPTO ODP "PTAB proceedings on file" block returns no AIA trial proceedings as of the most recent ingest. Per the task constraints, that structured list is canonical.
• Web corroboration: searches on the patent number, the inventors (Gallant, Naselaris, Kay, Prenger), and the assignee (Regents of the University of California) returned no petition or decision for this patent. Google Patents' page for US9451883B2 likewise shows no PTAB proceedings section.
• False-positive filter (important): search hits mentioning an "'883 patent" almost always refer to different patents. Sonos's '883 patent (a 2020-issued playback-device patent, subject to IPR2025-00509/00511 family petitions) and various Torgerson telemetry '883 references are unrelated to US 9,451,883 and must not be attributed to this patent.
• Related-file check: the priority chain is a provisional (61/157,310, filed 2009-03-04), a parent nonprovisional (12/715,557, filed 2010-03-02, now abandoned), and this CIP (13/725,893, filed 2012-12-21). No PTAB activity attaches to any of them either.

Strategic summary

Claim status — everything is untested. All claims of US 9,451,883 stand exactly as issued; none have been canceled, confirmed, or even construed by the Board. Because no FWD exists, there is no claim-level disposition to summarize. For reference, the patent's claim set as issued runs to at least claims 1–20, with independent claim 1 (method) and claim 13 (apparatus) reciting decoding via a decoding distribution over one or more linearizing feature spaces, and dependent claims adding specifics such as a Gabor wavelet non-linear transform and cognitive-state variants. Those are the claims a defendant would face — but their precise scope has never been adjudicated anywhere. (Note: the Office of the Chief Administrative Officer's public copies and third-party mirrors of the granted text are not perfectly consistent with the 2013 pre-grant publication US2013/0184558A1; pull the granted claims from the USPTO Patent Center PDF rather than from a mirror before charting anything.)

Estoppel landscape — empty. With no IPR/PGR, there is no § 315(e)(1) or § 315(e)(2) estoppel running against any party, and no petitioner or privy is boxed out of any ground. Conversely, there is also no § 325(e) estoppel binding the patent owner. If you file first, you would bear the full § 315(b) one-year bar and the § 315(e)(2) estoppel risk yourself, with no second-chance safety net — so any petition should be drafted to include every art combination you might want later, including § 112 and § 101 theories you would otherwise reserve for district court.

Pattern signals — none. No petitioner has filed even once, so there is no serial-petition or General Plastic pattern. The patent owner (UC Regents) has not had to defend a single PTAB appeal and has no Federal Circuit PTAB-appeal history on this patent. There is no defensive aggregator (Unified Patents, RPX, etc.) anywhere in the chain. Notably, the patent's status as an unasserted, university-owned research tool patent — with an infringement-theory problem as much as a validity problem, given that the claims require measuring brain activity with an fMRI/MEG/EEG-class device — is a plausible explanation for the absence of IPRs.

Recommended next steps

1. Verify the maintenance-fee status immediately — this is the highest-value check. The structured data's legal status is "Expired - Fee Related, expires 2030-04-20." That label typically reflects a lapsed maintenance fee. Pull the fee history on USPTO Patent Center (https://patentcenter.uspto.gov) for US 9,451,883. If a 7.5- or 11.5-year maintenance fee was missed, the patent may be unenforceable for intervening infringement subject to a § 41(c) petition for unintentional-delay reinstatement — which a patent owner can still file, so treat expiry as a strong but reversible defense. Do not represent to a court that the patent is expired without checking the current fee record.
2. If the patent is in force and you are accused, budget for a first-ever IPR. Because no PTAB record exists, there is no free-riding path. Grounds worth scouting: the Gabor-wavelet / motion-energy / semantic-category decoding literature (e.g., Kay, Naselaris, Prenger & Gallant, Neuron 2008; Mitchell et al. 2008; Nishimoto et al. 2011) is largely the inventors' own pre-2009 work, which makes it § 102(b)/(a) art against the 2009 priority date and also raises § 102(f)/inventorship- and derivation-adjacent questions. The Miyawaki et al. 2008 Neuron reference, already of record on the face of the patent, points at the closest non-inventor art.
3. Do not cite either of the following to a court: (a) any "'883 patent" FWD from the Sonos or telemetry proceedings (different patents), or (b) any statement that claims have been canceled — none have.
4. Statutory-deadline mechanics, if you file: institution decision is due within 6 months of the petition's filing date under § 314(b), and the FWD is due within 12 months of institution under § 316(a)(11) — there is no scheduled milestone to track today because no proceeding is pending.
5. Parallel-track check: confirm via CourtListener (https://www.courtlistener.com) and PACER whether US 9,451,883 has ever been asserted. No litigation asserting it surfaced in this search, but a stale or sealed docket would not necessarily appear; if it has been asserted, the § 315(b) one-year clock is running against those defendants and the first-filer advantage is live.

I'll research the assignment record for US 9451883 before writing anything. Let me start with several targeted searches.

Let me dig for the specific reel/frame record and any litigation involving this patent.

Ownership Reconstruction — US 9,451,883 B2

Patent: US 9,451,883 B2 — "Apparatus and method for decoding sensory and cognitive information from brain activity"
Application: 13/725,893 · filed 2012-12-21 · granted 2016-09-27 · published US20130184558A1 (2013-07-18)
Priority: 2009-03-04 (provisional 61/157,310); CIP of Ser. No. 12/715,557 filed 2010-03-02

Sourcing note / limitation up front. The authoritative Google Patents record (fetched 2026-09-28) shows one post-filing legal event of assignment type and no post-issuance transfers. However, I was not able to retrieve the reel/frame number or the correspondent of record for that entry — Google Patents' legal-events feed renders it as "ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS)" without the reel/frame, and my searches did not successfully hit the Assignment Center / assignment.uspto.gov record for this patent. I did not fabricate reel/frame numbers or correspondent names. Where a field could not be verified, it is marked [not retrieved]. Everything below that is asserted is grounded in the patent text and the legal-events record.

Inventors

Pattern note (unusual): These four are the same cohort behind the well-known "Bayesian Reconstruction of Natural Images from Human Brain Activity" (Neuron 2009) and "Identifying natural images from human brain activity" (Nature 2008) work. The inventions here are academic-lab inventions, not product-line inventions. I found no evidence of inventors departing an operating-company assignee within 12 months of filing (there was no operating-company assignee to depart from). The relevant caveat is instead a university → system-office nuance: the published work is Berkeley-affiliated, while Google Patents lists the original/current assignee as "University of California San Diego UCSD." Both are campuses of the same legal owner (the Regents of the University of California), so this is a same-owner designation, not a transfer — but the campus mismatch is worth flagging as a record-keeping artifact rather than a real ownership event.

Original assignee

Regents of the University of California, a California corporation (Google Patents also displays "University of California San Diego UCSD" as original assignee; per the 2013-03-26 recordation, the named assignee entity is the Regents of the University of California).

• Product embodying the claims: None. This is a research-methodology patent (fMRI/EEG/MEG brain-activity decoding and stimulus reconstruction). The Regents do not ship a commercial product reading these claims; the claims cover a decoding method and an apparatus (processor + executable programming) for reconstructing perceptual/cognitive content from measured brain activity.
• Primary line of business: Public research university system (education, sponsored research, technology transfer via its Office of Technology Transfer / campus tech-transfer offices).
• Current status: Operating (the Regents are an ongoing institution; not acquired, not dissolved, not in bankruptcy).
• Patent status: Expired – Fee Related, per Google Patents, with an adjusted expiration of 2030-04-20. That lapse-for-non-payment is significant: the maintenance fee was not paid, which is itself evidence of a non-asserting, non-monetizing posture for this asset.

Assignment timeline

Chronological list of every recorded assignment (as surfaced):

• 2013-03-26 (recorded) — Reel [not retrieved] / Frame [not retrieved]
  • Execution date: [not retrieved] (execution date not shown in the legal-events feed)
  • Conveyance: Assignment of Assignors' Interest ("see document for details")
  • Assignor: Naselaris, Thomas; Prenger, Ryan; Gallant, Jack L.; Kay, Kendrick (all four named inventors)
  • Assignee: Regents of the University of California, a California corporation
  • Correspondent: [not retrieved] — no attorney/agent of record captured in the sources I could reach. I flag this explicitly because the brief asks the correspondent be captured; I do not have it and will not invent it. (With only one assignment in the chain, the "repeat correspondent" tell is in any event unavailable.)
  • Context: Initial inventor-to-employer assignment — routine university-employment assignment of the inventors' rights to the Regents; not an acquisition, fire-sale, reorg, securitization, or transfer-to-asserter.

No other assignment records were found. There is no post-issuance transfer of any kind: no assignment to an IP-holding LLC, no security interest, no merger, no change of name, no license recordation, no release. There is also no recorded reassignment after grant and no litigation found naming US 9,451,883 in any district court or PTAB source I could reach.

Bottom line for this section: the Assignment Center / legal-events record for this patent contains only the original inventors→Regents assignment. Per the operating rules for this review, that is itself the finding: the Regents of the University of California still own the patent, subject to the recordation caveat below.

Timeline diagram

timeline
    title Ownership of US 9451883
    2009 : Priority provisional filed
    2010 : Parent application filed
    2012 : CIP application filed
    2013 : Inventors assign to UC Regents
    2016 : Patent issued
    2030 : Adjusted expiration date

NPE / troll-pattern signals

1. Shell-entity transfer — not present. There is no assignment to any "IP / Patents / Licensing / Holdings / Ventures" entity. The sole recorded assignee is the Regents of the University of California (a public university corporation), not a single-purpose Delaware or Texas LLC, and no registered-agent-service address appears.

2. Known asserter in the chain — not present. The only assignee (Regents of the University of California) does not appear on the Acacia / Marathon / Intellectual Ventures / IPNav / Wi-LAN / Conversant / Vringo / Pendrell / Innovatio / MPHJ / Lumen View / Round Rock / Erich Spangenberg-type lists, and I found no Unified Patents or RPX high-frequency-plaintiff listing for this patent or its owner.

3. Repeat correspondent across the chain — not present / indeterminate. With only one recorded assignment there is no recurrence to detect. The correspondent of record for the 2013-03-26 entry is [not retrieved], so I cannot confirm or refute this tell either way. Marked indeterminate rather than "present."

4. Cascading transfers — not present. Zero consecutive chained assignments; the chain is a single link (inventors → Regents) over a ~4-year filing-to-grant window with no subsequent movement.

5. Pre-litigation transfer — not present. No assignment occurs within 6 months before any infringement suit, because no infringement suit naming this patent was found.

6. Bankruptcy fire-sale — not present. The assignee is a public university system; no Chapter 7/11 proceeding, no portfolio auction, no record of sale in proceedings.

7. Privateering — not present. No operating company transferred this patent to an NPE, and no NPE is in the chain to assert on anyone's behalf.

8. Defensive aggregator (anti-NPE) — not present. The chain does not terminate at RPX, AST, LOT, Unified Patents, or OIN. It terminates at the original university assignee — so this is not an "inverse signal" neutralization; it is simply an unmonetized academic asset.

Note also the fee lapse: Google Patents shows legal status Expired – Fee Related with adjusted expiration 2030-04-20. A lapsed, unasserted, university-held neuroscience-methods patent is affirmative evidence of a non-asserting posture, not merely absence of evidence.

Verdict

Insufficient data — only the original assignment exists.

Justification: The entire recorded chain is a single link: the four named inventors (Naselaris, Prenger, Gallant, Kay) assigned to Regents of the University of California by an assignment recorded 2013-03-26 (conveyance: "Assignment of Assignors' Interest"; reel/frame [not retrieved]). There are no subsequent assignments, no LLC transfers, no known NPE assignee, no litigation, and the patent lapsed Expired – Fee Related with adjusted expiration 2030-04-20 — all of which point away from any NPE/troll pattern. Per the review rubric, a chain consisting of only the original assignment is scored Insufficient data rather than a positive NPE finding; the substantive conclusion is that the Regents of the University of California remain the owner and the patent is non-asserting. The one open field is the correspondent of record, which I could not retrieve and therefore did not characterize.

Verification link: USPTO Assignment Center — search by patent number 9451883 (also indexed at assignment.uspto.gov).

I'll search for the specific patent and its cited prior art references.

Let me retrieve the front-page "References Cited" section more completely.

I have partial front-page reference data. Let me retrieve the complete patent citation list and the claims for accuracy.

Prior Art Analysis — US Patent 9,451,883 B2

Search status note before anything else

I ran the requested number of searches for the literal identifier 9451883 (and its variants 9,451,883 / US9451883B2 / US 20130184558 A1, which is its own pre-grant publication). I want to be transparent up front: I could not retrieve the complete, verbatim "(56) References Cited" block from the granted patent's front page. The OCR of the front-page PDF (patentimages.storage.googleapis.com/.../US9451883.pdf) came back fragmented, and the authoritative full-text feed you supplied contains the specification and abstract but not the front-page citation block. I recovered part of it, plus a large block of the non-patent-literature citations. I am flagging the gap rather than silently filling it, because inventing citation numbers would violate your literal-identifier rule.

Everything below distinguishes (a) what I actually retrieved from (b) inference, and does not auto-correct any identifier.

1. USPTO-level record verification (literal identifier match)

This confirms the number resolves to the Gallant / Naselaris / Kay / Prenger brain-decoding patent — not to any of the near-miss numbers flagged in the prior "litigation" section (7,454,002; 9,479,973; 10,420,883; etc.). Those near-misses remain excluded.

Refinement to a prior-section statement (not a contradiction): the earlier summary estimated "at least 17 claims … four independent claims implied." The evidence I retrieved now indicates the granted patent has three independent claims — 1 (method), 13 (apparatus), and 21 (method) — with claims running to at least 21 (source: PatentForecast claim matrix). Meanwhile the pre-grant publication (US 2013/0184558 A1) shows independents at 1, 10, 19, 20. That divergence tells us the claims were substantially amended during prosecution — which is itself relevant, because the amendments were the applicant's response to whatever art the examiner cited. I still cannot certify the exact total claim count.

2. The "(56) References Cited" block — what I recovered

The front-page OCR surfaced the following fragments verbatim, in this order:

U.S. Patent Documents (published applications):

• US 2007/0032737 A1 — "Causevic et al." — 2/2007
• US 2008/0208072 A1 — "Fatem et al." — 8/2008

Non-Patent Literature (recovered, partial list): Miyawaki et al.; Pasley et al., Reconstructing Speech from Human Auditory Cortex, PLoS Biol 10(1): e1001251; Olman et al. (Vision Research 2004); DeYoe et al. (PNAS 1996); Haynes et al. (PNAS 2004); Tootell et al. (PNAS 1998); Salinas & Abbott (J Comput Neurosci 1994); Sereno et al. (Science 1995); Sereno & Huang (Nature Neurosci 2006); Simoncelli & Olshausen (Annu Rev Neurosci 2001); Thirion et al., Inverse retinotopy (Neuroimage 2006); Naselaris et al., Bayesian Reconstruction of Natural Images from Human Brain Activity, Neuron 2009; Naselaris et al., Encoding and decoding in fMRI (NeuroImage 2010); Nishimoto et al., Reconstructing Visual Experiences from Brain Activity Evoked by Natural Movies, Current Biology 2011; Huth et al., A Continuous Semantic Space Describes the Representation of … Object and Action Categories across the Human Brain, Neuron 2012; Cao et al. (Comput Stat Data An 1994); Engel (Neuron); Boynton & Finney (J Neurosci 2003); Hansen et al. (J Neurosci 2007); Larsson & Heeger (J Neurosci 2006); O'Toole et al. (J Cogn Neurosci 2005); Bullmore (Hum Brain Mapp 2001); Carlson et al. (J Cogn Neurosci 2003); Smith et al. (Cereb Cortex 2001); Hanson et al. (NeuroImage 2004); Daugman (J Opt Soc Am A 1985); Lee, T.S. (IEEE TPAMI 1996); Lee, T. (Hum Brain Mapp 2002); Furmanski & Engel (Nature Neurosci 2000); Langleben et al. (Hum Brain Mapp 2005); Davatzikos et al. (NeuroImage 2005); Haxby et al. (Science 2001); Haynes & Rees (Nature Neurosci 2005); Haynes & Rees (Nat Rev Neurosci 2006); Heeger et al.

Not recovered: the complete ordered "(56)" list; any Foreign Patent Documents; and the examiner's own markings. Treat the above as partial.

3. Preliminary § 102 (anticipation) assessment per reference

Framing caveat (important)

On the face of the record, these are overwhelmingly § 103 / background references, not § 102 anticipatory references. For a reference to anticipate under § 102 it must disclose every element of a claim, arranged as in the claim. The granted independent claims require a combination: (i) two brain-activity acquisitions; (ii) conversion of the first into encoding-model parameter values for a linearizing feature space; (iii) a decoding database of candidate items; (iv) generation of predicted brain-activity signals per candidate; (v) a probability computation per candidate; (vi) selection of candidate(s); and (vii) reconstruction from the selection. No single reference retrieved discloses all of that. My honest assessment is therefore that no retrieved reference plainly anticipates, and the strongest § 102 posture belongs to the inventors' own publications against the later-filed (CIP) claims — an issue I address in § 4.

That said, here is the mandated per-reference mapping, with the claim-mapping keyed to granted independent claims 1, 13, 21 and the notable dependents (2/imaging Markush; 8 vision encodings Gabor/scene/motion-energy/WordNet; 9 auditory encodings spectral/phonemic/syntactic/semantic).

Bottom line on § 102: Among the references I could verify, none anticipates the full scope of granted independent claims 1, 13, or 21, because each omits at least the decoding-database/candidate-generation/selection pipeline that the claims require. The strongest § 102 exposure is Naselaris 2009 (#3), and it is exposure that arises only if the challenged claims are not entitled to the 2009 priority date.

4. The critical-date issue that governs the whole § 102 analysis

This is the single most important point, and it is easy to miss:

• Claims get the 2009-03-04 priority date only if the '557 application supports them. Because '883 is a continuation-in-part, any claim element that first appeared as new matter in the 2012-12-21 CIP is entitled only to the 2012-12-21 filing date. The decoding-database / candidate-item / probability-and-select architecture — which is prominent in the granted independent claims (esp. claims 13 and 21) — is the kind of language that commonly reflects CIP-added matter.
• Consequence: if claims 13/21 (and their dependents) are limited to 2012-12-21, then all four lab publications — Naselaris 2009, Naselaris 2010, Nishimoto 2011, and Huth 2012 — predate the effective filing date and become available as § 102(a)/(b) prior art (subject to any applicable one-year grace period and the fact that they are the inventors' own disclosures).
• If instead the claims get the 2009-03-04 date, those publications post-date priority and are not § 102 prior art (they'd be, at most, § 102(e)/(a) considerations and, being the inventors' own work, generally not anticipatory).

This is the dividing line between "clean patent" and "§ 102 exposure," and I cannot resolve it without the full granted claims and the '557 specification.

5. What this means, stated plainly

1. The patent's own cited art (the "(56)" list) is directional, not dispositive. It reads as a typical academic-patent citation block: the Gallant lab's own papers plus the neuroscience/statistics literature underlying the embodiments. That is background + § 103 material.
2. The genuinely material references are the lab's own four papers (Naselaris 2009; Naselaris 2010; Nishimoto 2011; Huth 2012) and Pasley 2012, because they map one-for-one onto the four independent-claim limitations and the dependent-claim encoding-model recitations (Gabor/scene = Naselaris 2009; motion-energy = Nishimoto 2011; WordNet = Huth 2012; auditory = Pasley 2012).
3. Whether those references are "prior art" at all depends on the CIP priority determination in § 4 — not on whether they disclose the invention (they plainly do).

6. Uncertainty disclosures (please read before relying)

• Incomplete source. The full verbatim "(56) References Cited" block (all U.S. patent documents, all foreign documents, and examiner categories) was not retrievable; the two U.S. published applications (Causevic; Fatem) and the NPL list above are what the OCR yielded. I did not fabricate any citation numbers to fill gaps.
• Claim text. My claim-element mapping uses the granted independent claims 1, 13, 21 as rendered by PatentForecast and the dependent claim text from the pre-grant publication (Justia). I have not verified the granted claims verbatim. Claim numbering changed between publication and grant (independents 1/10/19/20 → 1/13/21), so the mapping is provisional.
• § 102 vs § 103. I have been conservative: I identified anticipation candidates rather than asserting anticipation. Any conclusion that a reference anticipates requires element-by-element claim comparison against the granted claim text — a step I could not fully complete here.
• No auto-correction caveat. Every identifier above is reproduced literally. Where a citation's full bibliographic details were not recoverable (notably the two U.S. published applications), I have said so explicitly rather than reconstructing a title or number.
• Search budget. My search budget for this task was exhausted; a fuller result would require pulling the granted patent's front page directly from USPTO Patent Center / Patent Public Search and the file wrapper (which would also reveal the examiner's § 102/§ 103 rejections and the applicant's amendments — the most direct evidence of what art actually threatened these claims).

I'll start by verifying what the "Prior Art" section of the source page actually contains, then locate and date-check the real-world prior-art references that matter for a § 103 analysis of this 2009-priority patent.

Excellent — I've now recovered the actual "References Cited" list that constitutes the Prior Art section. Let me pin down the two cited US patent applications and the key non-patent references.

I ran out of search budget mid-verification, so I'll flag exactly what is verified vs. inherited. Here is the § 103 analysis.

Obviousness Analysis — US 9,451,883 B2 under 35 U.S.C. § 103

0. What the "Prior Art" section actually contains (and how I got it)

The Google Patents text feed passed to me did not include a citations block — it carried only machine-generated "Prior art keywords" (brain activity; stimuli; brain; activity data; stimulus) and a "Prior art date" of 2009-03-04. I therefore recovered the actual "(56) References Cited" list from the patent's own front page via the USPTO publication PDF and the Justia mirror:

• https://patentimages.storage.googleapis.com/d9/05/28/f1f8459d71a05f/US9451883.pdf
• https://patents.justia.com/patent/[9451883](/patent/9451883)

Recovered "References Cited" (partial — the feed truncated both ends of the list):

Identifier-discipline notes (I am not auto-correcting):

• The Huth reference as printed on the patent reads "Theories of Object and Action Categories." The published Neuron title is "Thousands of Object and Action Categories." I am recording the discrepancy rather than silently fixing the patent's text. Either way the citation is to Neuron vol. 76, pp. 1210–1224, Dec. 20, 2012.
• I did not find Kay, Naselaris, Prenger & Gallant, Identifying natural images from human brain activity, Nature 452:352–355 (Mar. 20, 2008) in the portion of the cited list I recovered. The recovered excerpt is partial, so I cannot state it is absent from the face of the patent — but it is squarely § 102(b) art relative to the 2012 CIP and is discussed below on that footing.
• "Pasley et al." is printed without a year; PLoS Biol 10(1):e1001251 corresponds to Jan. 31, 2012.

1. Explicit contradiction flag vs. the prior sections

The previously generated summary stated the granted claim set was truncated and reconstructed from the 2013 pre-grant publication, yielding a 20-claim set with claim 1 reciting "converting … into a corresponding set of predicted response values." That is consistent with the application publication (US 2013/0184558 A1).

However, the PatentForecast record for the granted patent reports 29 claims with materially different independents:

• Claim 1 now requires "converting … into a corresponding set of encoding model parameter values for each of one or more linearizing feature spaces," plus "creating a decoding database comprising one or more items," "generating … predicted brain activity signals for each … item," "selecting at least one item," and "producing a reconstructed set of brain activity stimuli based on the selected at least one item."
• Claim 13 is the apparatus mirror; claim 21 is a third independent reciting training stimuli → training feature spaces → "encoding model parameter values that model a linear relationship" → measured data from "at least one second subject" → candidate stimulus items → predicted activity → comparison → selection → reconstruction.

(Source: https://sectors.patentforecast.com/patent-matrix/US9451883)

These two claim sets cannot both be the granted claims. The full-text feed I was given reproduces the application claim set under the heading "Claims." Per the operating rule that the authoritative full text governs, I treat the 20-claim application set as what the page contains, but I must flag that the granted claim set (29 claims, per PatentForecast) is broader/different in the ways above, and that the § 103 analysis therefore has to be run on both. All mappings below are keyed to claim language common to both, and I say so where the granted text adds an element.

2. The dispositive threshold issue: the effective filing date

Pre-AIA §§ 102/103 govern (the application was filed 2012-12-21, before the March 16, 2013 AIA date). The chain is:

61/157,310 (2009-03-04) → 12/715,557 nonprovisional (2010-03-02) → 13/725,893 CIP (2012-12-21).

This matters more than any reference combination, because the patent's own front page cites Nishimoto 2011, Pasley 2012, and Huth 2012 — all of which postdate the 2009 provisional and two of which postdate the 2010 nonprovisional. Their presence in the cited list is consistent with the Office treating them as available prior art against at least some claims, i.e., as evidence that those claims were not accorded the 2009-03-04 date.

Concretely:

The specification content that only the CIP plausibly supports — the MPS/2000-channel spectral model, the 39-phoneme model, the 36-part-of-speech model, the LSA semantic model, the WordNet/1705 object-and-action-category model, and the motion-energy model with temporal FIR filters — is exactly the content of Naselaris 2011, Nishimoto 2011, Pasley 2012, and Huth 2012. That is a powerful internal-consistency argument that claims 8, 9, 17, 18, 26 and the auditory/semantic dependents get 2012-12-21 at the earliest, which makes the cited references anticipatory-flavored § 103 art.

Caveat: a same-inventive-entity reference is not § 102(a)/(b) art. Gallant, Naselaris, Kay and Prenger are the named inventors; but Kay 2008 adds no co-author, Naselaris 2009 adds Oliver, Nishimoto 2011 adds Orban, Huth 2012 adds Huth and Vu. Those are "by another" in part and therefore potentially § 102(a)/(b) art subject to a Rule 131 showing of prior invention. I cannot resolve inventorship-driven disqualification from the record available to me.

3. Elements-in-common across all three independents, and the art that discloses each

4. Combinations that render the claims obvious

Ground 1 — Miyawaki 2008 + Thirion 2006 + Salinas & Abbott 1994 → core decode-and-reconstruct

Miyawaki discloses fMRI-based visual image reconstruction by combining multiscale local image decoders: fit local decoders on measured responses, apply them to candidate image elements, and reconstruct. Thirion discloses inversion of retinotopic response models to infer image content. Salinas & Abbott supplies the linear weighting/reconstruction theory.

Why combined: identical field (fMRI decoding of visual content), identical problem (recover the stimulus, not its category). Both teach away from classification by using continuous image parameters. Result: claim 1's (a)–(c), (f), (h) at minimum, and the 20-claim version's claim 1 in its entirety.

Ground 2 — Miyawaki 2008 + Kay 2008 + Naselaris 2009 → the "linearizing feature space + per-voxel encoding model + database scoring" architecture

Kay 2008 is the express motivation: it states that its receptive-field-model decoder "suggest[s] that it may soon be possible to reconstruct a picture of a person's visual experience from measurements of brain activity alone." Naselaris 2009 supplies precisely the patent's architecture: per-voxel encoding models, combined into a multivoxel model, inverted to a posterior over a large image database, with a prior (including a Monte Carlo empirical prior over six million images) — which is the "decoding distribution" of claim 6/15 and the "decoding database" + "probability for each item" of granted claim 1.

Why combined: Kay and Naselaris are the same research line; Naselaris 2009 itself cites and builds on Kay 2008. There is a finite, identified set of predictable solutions (encoding-model inversion), KSR‑sanctioned.

Ground 3 — Naselaris 2009 + Nishimoto 2011 (+ Nishimoto's database-from-the-internet approach) → "decoding database," time-varying stimuli, and top-N aggregation

Nishimoto 2011 extends reconstruction from static images to natural movies using a motion-energy feature space and reconstructs by matching measured activity against a large library of natural video, selecting the best-matching excerpts. That is the "creating a decoding database [of] items … selecting at least one item … producing a reconstructed set of stimuli" of granted claim 1, and it independently supports claim 4 (plurality of selected items; average/weighted average) because a database search returns ranked candidates, not a unique match — and the patent's own specification concedes exactly this ("The best match does not need to be selected here. Instead, the top ten, one-hundred, etc. items … can be averaged, or a weighted average may be taken").

Ground 4 — Huth 2012 (+ Naselaris 2009 semantic model; Mitchell 2008 semantic-feature construction) → semantic feature spaces and claims 9/18, 27

Naselaris 2009 already used a semantic encoding model (23 human-labeled categories) fused with the structural/Gabor model — i.e., two linearizing feature spaces. Huth 2012 generalizes this to a continuous semantic space spanning ~1705 object and action categories derived from the WordNet hierarchy — which is verbatim the patent's WordNet encoding model (1705 categories, bin(W·l), identity-matrix rows). Mitchell 2008 supplies the earlier, well-known technique of building semantic feature spaces from text-corpus statistics and predicting fMRI patterns for words.

Why combined: same objective (capture semantic content that the Gabor/structural model cannot), same modality, and Huth is the direct successor to Naselaris 2009. Combination is a predictable, confirmable improvement, satisfying KSR.

Ground 5 — Pasley 2012 (+ Mitchell 2008 / Naselaris 2011) → auditory claims 9/18

Pasley 2012 reconstructs speech from human auditory cortex using spectrogram-based decoding. A POSITA seeking to extend the framework of Grounds 1–2 to audition would apply the same encoding-then-invert architecture to a spectral feature space — precisely the patent's spectral/MPS model. Nishimoto and Naselaris 2011 together teach that the framework is modality-general.

Ground 6 — Causevic 2007/0032737 and/or Fadem 2008/0208072 → apparatus claims 13, 20, and the imaging-modality dependents (2, 11)

Both are programmed-computing-device EEG/bipotential analysis systems. Once the method steps are obvious, reciting "a processor and programming executable on the processor" to perform them is the routine software-implementation of a known method and adds nothing patentable. The Markush group EEG/MEG/fMRI/fNIRS/SPECT/ECoG is a closed list of all known brain-measurement modalities — enumerated alternatives, per se obvious where the choice is a matter of known capability and cost (a trade-off the patent's own specification concedes: "Some methods … such as EEG are cheap and convenient, but they have low signal quality. Others, such as fMRI, are of high quality, but they are expensive and inconvenient").

5. Motivation to combine (KSR factors)

1. Same field of endeavor and same problem. Every substantive reference addresses non-invasive decoding of brain activity; several expressly criticize the category-classification paradigm and propose reconstruction instead (Kay 2008; Naselaris 2009; Miyawaki 2008).
2. Express lead in the art. Kay 2008 explicitly forecasts the reconstruction that the patent claims.
3. Predictable result / reasonable expectation of success. Naselaris 2009 and Nishimoto 2011 actually produced reconstructions of novel natural images and movies. The combination (feature space + per-voxel linear encoding model + database inversion) had already been reduced to practice by 2011.
4. Finite number of identified, predictable solutions. Bayesian inversion of a linear encoding model with an empirical stimulus prior; adding a second feature space; using a larger candidate corpus — each is a small, enumerated set.
5. Design incentives. Larger stimulus databases and additional feature spaces demonstrably improve reconstruction quality (Naselaris 2009 shows priors "have a substantial effect on the quality of natural image reconstructions"), giving an explicit motivation to (i) build a decoding database, (ii) use multiple linearizing feature spaces, and (iii) average/weight several candidates.
6. No teaching away. The patent disparages classifiers; the cited art does not disparage encoding-model inversion — it teaches it. The patent's asserted advance (novel/untrained stimuli; reconstruction rather than classification) is the prior art's own stated goal.
7. Modality transferability. Once demonstrated for vision and audition separately, a POSITA would expect the same architecture to work for motor/cognitive stimuli and for other measurement modalities — supporting claims 3, 5, 10, 12, 20.

6. The patentee's strongest rebuttals, and my assessment

7. Bottom line

• If the claims receive the 2012-12-21 CIP date: the cited prior art on the patent's own face renders the independents (1, 13, 21) and most dependents prima facie obvious — chiefly Naselaris 2009 + Nishimoto 2011 + Miyawaki 2008, with Huth 2012 supplying the WordNet/semantic feature space and Pasley 2012 the auditory feature space.
• If the claims receive the 2010-03-02 date: Miyawaki 2008 + Kay 2008 + Naselaris 2009 still make the core decode/reconstruct architecture obvious, but the movie/auditory/WordNet dependents fall back on the CIP date anyway.
• If some claims receive 2009-03-04: Ground 1 (Miyawaki 2008 + Thirion 2006 + Salinas & Abbott 1994) is the surviving attack, and it targets only the generic decode-and-reconstruct concept — which is exactly what the 20-claim set's claim 1 recites.

Confidence: High on the reference identities, dates of publication for Kay 2008 / Naselaris 2009 / Miyawaki 2008 / Mitchell 2008 / Nishimoto 2011 / Huth 2012 / Pasley 2012; high on the element mapping; low-to-moderate on which claim set is the operative granted set, and therefore on precisely which limitations must be met. Unverified: the complete cited list (both ends truncated), the exact granted claim count and text for claims 2–12 and 14–20 of the 29-claim set, whether Kay 2008 appears on the face of the patent at all, and any Rule 131/priority determinations in the file history. I could not reach PACER, the PTAB, or the full prosecution history within my search budget. A definitive § 103 opinion needs the file wrapper and a claim-by-claim written-description support chart against 61/157,310 and 12/715,557.